Proceeding of the 2004 American Control Conference WeA13.

1
Boston, Maasachusens June 30 -July 2,2004

A System-Theoretic Foundation for

Thermodynamics: Energy Flow, Energy Balance,
Energy Equipartition, Entropy, and Ectropy
Wassim M. Haddadt, VijaySekhar Chellaboina', a n d Sergey G. Nersesovt

t School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0150

'Mechanical and Aerospace Engineering, University of Missouri, Columbia, MO 65211-2200

Absfracl-Thermodynamics is a physical branch of science not restrict the ability to convert work into heat or heat
that governs the thermal behavior of dynamical systems into work, except that energy must be conserved in the
as simple as refrigerators io as complex as our expanding process, The second law of thermodynamics asserts that
universe. The development of thermodynamics spawned out while the system energy IS always conserved, it will be
of steam tables and venous bleeding with many scientists degraded to a point where it cannot produce any useful
and engineers expressing concerns about the completeness work. Hence, it is impossible to extract work from heat
and clarity of its mathematical exposition over its tortuous without at the same time discarding some heat giving rise
history. In this paper we develop a system-theoretic founda- to a monoronically increasing quantity known as entropy.
tion Cor thermodynamics using n large-scale dynamical sys- While energy describes the state of a dynamical system,
tems perspective. Specifically, using compartmental dynamical entropy refers to changes in the sfatus quo of the system and
system energy Bow models, we place the universal energy is a measure of molecular disorder and the amount of wasted
conservation, energy equipartition, temperature equipartition, energy in a dynamical (energy) transformation from one
and entropy nonconservation laws of thermodynamics on
a system-theoretic foundation. Furthermore, we introduce state (form) to another. Since the system entropy monotoni-
a new and dual notion to entropy; namely, ecrropy, as a cally increases, the entropy of the dynamical system tends to
measure of the tendency of a dynamical system to do useful a maximum and thus time, as determined by system entropy
work and show that conservation of energy in an isolated increase [ 1-31, flows on in one direction only. Even though
thermodynamic system necessarily leads to nonconservation entropy is a physical property of matter which is not directly
of ectmpy and entropy. I n addition, using the system ectropy observable, it permeates the whole of nature, regulating
as a Lyapunov function candidate we show that our large-scale the nrmw of rime and responsible for the enfeeblement
thermodynamic energy flow model has convergent trajectories and eventual demise of the universe. While the laws of
to Lyapunov stable equilibria determined by the large-scale thermodynamics form the foundation to basic engineering
system initial subsystem energies. systems as well as nuclear explosions, cosmology, and our
expanding universe, many engineers and scientists have
expressed concerns about the completeness and clarity of
I. INTRODUCTION the different expositions of thermodynamics over its long
and flexuoils history, see r4-121.
Energy is a concept that underlies our understanding of Since the specific motion of every molecule of a ther-
all physical phenomena and is a measure of the ability of a modynamic system is impossible to predict, a mncroscopic
dynamical system to produce changes (motion) in its own model of the system is typically used with appropriate
system state as well as changes in the system states of macroscopic states which include pressure, volume, temper-
irs surroundings. Thermodynamics is a physical branch of ature, internal energy, and entropy, among others. However,
science that deals with laws governing energy Row from one a them>odynamically consistent energy Aow model should
body tu another and energy transformations from one form ensure that the system energy can be modelled by a dif-
to another. These energy flow laws are captured by the fun- fusion (conservation) equation in the form of a parabolic
damental principles known as the first and second laws of partial differential equation. These systems are infinite-
thermodynamics. The first law of thermodynamics gives a dimensional and hence finite-dimensional approximations
precise formulation of the equivalence of heat and work and are of very high order giving rise to large-scale dynamical
states that among all system transformations, the net system systems. Since energy is a fundamental concept in the
energy is conserved. Hence, energy cannot be created out of analysis of large-scale dynamical systems and heat (energy)
nothing and cannot be destroyed, merely transferred from is a fundamental concept of thermodynamics involving
one form to another. The law of conservation of energy the capacity of hot bodies (more energetic subsystems) to
is not a mathematical truth, but rather the consequence produce work, thermodynamics is a theory of large-scale
of an immeasurable culmination of observations over the dynamical systems. High dimensional dynamical systems
chronicle of our civilization and is a fundamental uxiom of can arise from both macroscopic and microscopic points
the science of heat. The first law does not tell us whether of view. Microscopic thermodynamic models can have the
any particular process can actually occur; that is, it does form of a distributed parameter model or a large-scale sys-
tem model comprised of a large number of interconnected
This research was supponed in pan by AFOSR under Grant F49620- subsystems. In contrast to macroscopic models involving
03-1-0178 and NSF under Grant ECS-0133038.

0-7803-8335-4/04/$17.00 02004 AACC 396

the evolution of global quantities (e.g., energy, tempera- arise from second-moment analysis of state space systems
ture, entropy, etc.), microscopic models are based upon under the assumption.of weak coupling. Even though these
the modeling of local quantities that describe the atoms results can be potentially applicable to linear large-scale
and molecules that make up the system, and their speeds, dynamical systems with weak coupling, such connections
energies, masses, angular momenta, behavior during colli- are not explored in [23]. With the notable exception of [34],
sions, etc. The mathematical formulations based on these none of the aforementioned SEA-related works address the
quantities form the basis of stafisfical mechanics. Since second law of thermodynamics involving entropy notions
microscopic details are obscured on the macroscopic level, in the energy flow between subsystem.
it is appropriate to view a microscopic model as an inherent The goal of the present paper is directed toward placing
model of uncertainty. However, for a thermodynamic system thermodynamics on a system-theoretic foundation. Specif-
the macroscopic and microscopic quantities are related since ically, since thermodynamic models are concerned with
they are simply different ways of describing the same energy flow among subsystem, we develop a nonlinear
phenomena. Thus, if the global macroscopic quantities can compartmental dynamical system model that is character-
be expressed in terms of the local microscopic quantities, ized by energy conservation laws capturing the exchange
the laws of thermodynamics could be described in the of energy between coupled macroscopic subsystems. Fur-
language of statistical mechanics. This interweaving of thermore, using graph theoretic notions we state two ther-
the microscopic and macroscopic points of view lead to modynamic axioms consistent with the zeroth and second
diffusion being a natural consequence of dimensionality laws of thermodynamics that ensure that our large-scale
and, hence, uncertainty on the microscopic level despite the dynamical system model gives rise to a thermodynamically
fact that there is no uncertainty about the diffusion process consistent energy flow model. Specifically, using a large-
per se. scale dynamical systems theory perspective for thermody-
In the last half of the 20th century thermodynamics namics, we show that our compartmental dynamical system
was re-formulated as a global nonlinear field theory with model leads to a precise formulation of the equivalence
the ultimate objective to determine the independent field between work energy and heat in a large-scale dynamical
variables of this theory [13-151. This aspect of thermody- system. Next, we give a deterministic definition of entropy
namics, which became known as rational thermo+namics, for a large-scale dynamical system that is consistent with the
was predicated on an entirely new axiomatic approach. As a classical thermodynamic definition of entropy and show that
result of this approach, modem continuum thermodynamics it satisfies a Clausius-type inequality leading to the law of
was developed using theories from elastic materials, viscous entropy nonconservation. Furthermore, we introduce a new
materials, and materials with memory [16-19]. Connections and dual notion to entropy; namely, ectropy, as a measure of
between thermodynamics and system theory as well as in- the tendency of a large-scale dynamical system to do useful
formation theory were also explored [2&27]. For an excel- work and show that conservation of energy in an isolated
lent exposition of these different facets. of thermodynamics thennodynamically consistent system necessarily leads to
see 1281. Thermodynamic principles have also been repeat- nonconservation of ectropy and entropy. Then, using the
edly used in coupled mechanical systems to arrive at energy system ectropy as a Lyapunov function candidate we show
flow models with modal energy playing the role of tem- that our thermodynamically consistent large-scale nonlinear
perature. Specifically, in an attempt to approximate high- dynamical system model possesses a continuum of equilib-
dimensional dynamics of large-scale structural (oscillatory) ria and is semistable; that is, it has convergent subsystem
systems with a low-dimensional diffusive (non-oscillatory) energies to Lyapunov stable energy equilibria determined by
dynamical model, structural dynamicists have developed the large-scale system initial subsystem energies. In addi-
thermodynamic energy flow models using stochastic energy tion, we show that the steady-state distribution of the large-
flow techniques. In particular, statistical energy analysis scale system energies is uniform leading to system energy
(SEA) predicated on averaging system states over the statis- equipartitioning corresponding to a minimum ectropy and a
tics of the uncertain system parameters has been extensively maximum entropy equilibrium state. In the case where the
developed for mechanical and acoustic vibration problems subsystem energies are proportional to subsystem tempera-
[29-341. Thermodynamic models are derived from large- tures, we show that our dynamical system model leads to
scale dynamical systems of discrete subsystems involving temperature equipartition wherein all the system energy is
stored energy flow among subsystem based on the as- transferred into heat at a uniform temperature. Furthermore,
sumption of weak subsystem coupling or identical subsys- we show that our system-theoretic definition of entropy and
tems. However, the ability of SEA to predict the dynamic the newly proposed notion of ectropy are consistent with
behavior of a complex large-scale dynamical system in Boltzmann’s kinetic theory of gases involving an n-body
terms of painvise subsystem interactions is severely limited theory of ideal gases divided by diathermal walls.
by the coupling strength of the remaining subsystems on The contents of the paper are as follows. In Section I1
the subsystem pair. Hence, it is not surprising that SEA we establish notation, definitions, and review some basic
energy flow predictions for large-scale systems with strong results on nonnegative and compartmental dynamical sys-
coupling can be erroneous. Altematively, a deterministic tems. In Section 111 we use a large-scale dynamical sys-
thermodynamically motivated energy flow modeling for tems perspective to provide a system-theoretic foundation
structural systems is addressed in [35], [36]. This approach for thermodynamics. Specifically, we develop a nonlinear
exploits energy flow models in terms of thermodynamic compartmental dynamical model characterized by energy
energy (i.e., ability to dissipate heat) as opposed to stored conservation laws that is consistent with the basic thermo-
energy and is not limited to weak subsystem coupling. dynamic principles. Then we turn our attention to stability
Finally, a stochastic energy flow compartmental model (i.e., and convergence. In particular, using the total subsystem
a model characterized by conservation laws) predicated on energies as a candidate system energy storage function,
averaging system states over the statistics of stochastic
system exogenous disturbances is developed in [23]. The we show that our thermodynamic system is lossless and
basic result demonstrates how linear compartmental models hence can deliver to its surroundings all of its stored
subsystem energies and can store all of the work done to
397
all of its subsystems. Next, using the system ectropy as a (respectively, positive) if W(i,j) 2 0 (respectively, W(i,j) >
Lyapunov function candidate we show that the proposed O), i , j = 1,.. . ,q.
thermodynamic model is semistable with a uniform energy The following definition introduces the notion of essen-
distribution corresponding to a minimum ectropy and a tially nonnegative functions [24], [38].
maximum entropy. In Section IV we generalize the results
of Section I11 to the case where the subsystem energies in Definition 2.2; Let w = [wlr. . . ,wqlT : V + Rq, where
large-scale dynamical system model are proportional to sub- V is an open subset of that contains E'+:Then w is
system temperatures and arrive at temperature equipartition essentially nonnegative if wi(z) 2 0 for all z = 1,.. . ,q
for the proposed thermodynamic model. Furthermore, we and t E E: such that ti = 0, where ti denotes the ith
provide a kinetic theory interpretation of the steady-state component o f t .
expressions for entropy and ectropy. In Section V, we spe-
cialize the results of Section I11 to thermodynamic systems Note that if ~ ( = t W ) z , where W E W9'q, then w(.)
with linear energy exchange. In Section VI we extend the is essentially nonnegative if and only if W is an essentially
results of Section I11 to continuous thermodynamic systems nonnegative matrix.
wherein the subsystems are uniformly distributed over an n
dimensional (not necessarily Euclidian) space. Specifically,
Proposition 2.1 ([24]. [38]): Suppose z+c V . Then
we develop a nonlinear distributed parameter model wherein W, .is an invariant set with respect to
-4

the system energy is modeled by a diffusion (conservation)

equation in the form of a parabolic partial differential i ( t )= w(.(t)), +U) = 20, t 2 to, (1)
equation. Energy equipartition and semistability are shown
using the well-known Sobolev embedding theorems and the where to E E:, if and only if w : V + R' is essentially
notion of generalized (or weak) solutions. Finally, we draw nonnegative.
conclusions in Section VII. The following corollary to Proposition 2.1 is immediate.
Corollary 2.1: Let W E W X q . Then W is essentially
11. MATHEMATICAL
PRELIMINARIES nonnegative if and only if ewt is nonnegative for all t 2 0.
The following definition introduces several types of sta-
In this section we introduce notation, several definitions, bility for the nonnegative dynamical system (1).
and some key results needed for developing the main
results of this paper. Let W denote the set of real numbers, Definition 2.3: The equilibrium solution t ( t ) = te of
W" denote the set of n x 1 column vectors, ( , ) T denote (1) is Lyapunov stable if, for every E > 0, there exists
transpose, (.)# denote group generalized inverse, and let 6 = a(€) > 0 such that if to E Da(z,) n E:, then z(t) E
I, or I denote the n x n identity matrix. For U E Wq D,(t,) n E:, t 2 to. The equilibrium solution z(t) G ze
we write U 2 2 0 (respectively, U >> 0) to indicate that of (1) is semistable if it is Lyapunov stable and there exists
every component of U is nonnegative (respectively,positive). 6 > 0 such that if to E Da(z,) nk;, then limt,,t(t)
In this case we say that U is nonnegative or positive, exists and converges to a Lya unov stable equilibrium point.
respectively. Let E: and R$ denote the nonnegative and The equilibrium solution t(ty t. of (1) is asymptotically
positive orthants of Wq; that is, if U E Wq, then w E E: stable if it is Lyapunov stable and there exists d > 0
and U E RP, are equivalent, respectively, to U 2 2 0 and such that if to E Ba(r,) n , :E then limt,, t ( t ) = 2.,
U >> 0. Furthermore, let i5S and denote the boundary Finally, the equilibrium solution z ( t ) = t, of (1) is globally
and the closure of the set S, respectively. Finally, we write asymptotically stable if the previous statement holds for all
11 . 11 for the Euclidean vector norm, /I . 115 for the operator to E E;.
norm of an element in a Banach space D, R ( M ) for the Finally, recall that a matrix W E W q x q is semistable
range space of a matrix M , V'(z) for the Frkchet derivative if and only if limt,, ewt exists [23], [24] while W is
of V at z, &(a), a E W", E > 0, for the open ball asymptotically stable if and only if limt,, ewt = 0.
centered at a with radius E, M 2 0 (respectively,A4 > 0) to
denote the fact that the Hermitian matrix M is nonnegative
(respectively, positive) definite, and z ( t ) + M as t + w
to denote that z ( t ) approaches the set M ; that is, for each 111. A SYSTEM-THEORETIC
FOUNDATION FOR
E > 0 there exists T > 0 such that dist(z(t),M) < E THERMODYNAMICS
for all t > T , where dist(p,M) 4 inf,,M Ilp - ~ 1 ( ~ , The fundamental and unifying concept in the analysis
The following definition introduces the nohon of Z-, M - of complex (large-scale) dynamical systems is the concept
, essentially nonnegative, compartmental, and nonnegative of energy. The energy of a state of a dynamical system
matrices. is the measure of its ability to produce changes (motion)
Delinition 2.1 ([23]. (241, [37]): Let W E Rqxq. W is in its own system state as well as changes in the system
a Z-matrix if W(i,3)5 0, i , j = 1,.. . , q, i # j. W is an states of its surroundings. These changes occur as a direct
M-matrix (respectively, a nonsingular M-matrix) if W is a consequence of the energy flow between different subsys-
Z-matrix and all the principal minors of W are nonnegative tems within the dynamical system. Since heat (energy) is a
(respectively, positive). W is essentially nonnegative if -W fundamental concept of thermodynamics involving the ca-
is a 2-matrix; that is, W 2 0, i , j = 1 , .. . ,q, i # j . pacity of hot bodies (more energetic subsystems) to produce
W is compartmental if d'
'IS essentially nonnegative and
W(i,j) 5 0, j = 1,...,q. Finally, W is nonnegative'
work, thermodynamics is a theory of large-scale dynamical
systems. As in thermodynamic systems, dynamical systems
can exhibit energy that becomes unavailable to do useful
'In this paper i t is important to distinguish behvween a square nonnegative work. This in tum contributes to an increase in system
(respectively. positive) matrix and a nonnegative-definite (respectively, entropy; a measure of the tendency of a system to lose the
positive-definite) matrix. ability to do useful work. In this section we use a large-scale
398
 Note that (2) yields a conservation of energy equation and
implies that the energy stored in the ith subsystem is equal
to the external energy supplied to (or extracted from) the
ith subsystem plus the energy gained by the ith subsys-
tem from all other subsystems due to subsystem coupling
minus the energy dissipated from the ith subsystem to the
environment. Equivalently, (2) can be rewritten as

&(t) = IUij(K(t)) - .ji(K(t))l - U i i ( K ( t ) )

j=l,j#i
+u(t), % ( t o ) = vsio, t 2 to, (5)
or, in vector form,
k(t) = W(v,(t))- d(v,(t)) + S ( t ) ,v,(to) = KO,
t 2 to, (6)
where Vso 2 [wale, ...,vsqo] T , yielding a power balance
equation that characterizes energy flow between subsystems
of the large-scale dynamical system D. Equation ( 5 ) shows
_ _ _ _ _ ~ _ _ ~ - - _ - - _ -
that the rate of change of energy, or power, in the ith
subsystem is equal to the power input (heat flux) to the ith
Fig. I . Large-scale dynamical system D subsystem plus the energy (heat) flow to the ith subsystem
from all other subsystems minus the power dissipated from
the ith subsystem to the environment. Note that (3) or,
equivalently, (6) is a statement of the first law of ther-
dynamical systems perspective to provide a system-theoretic modynamics for each of the subsystems with u s * ( , ) ,si(.),
foundation for thermodynamics. u i j ( . ) ,i # j, and ui;(.), i , j = 1,...,q, playing the role of
To develop a system-theoretic foundation for thermody- the zth subsystem internal energy, rate of heat supplied to
namics, consider the large-scale dynamical system D shown (or extracted from) the ith subsystem, heat flow between
in Figure 1 involving energy c c h a n e between q intercon- subsystems due to coupling, and the rate of energy (heat)
nected subsystems. Let vgi : R+ --t R, denote the energy dissipated to the environment, respectively. To further elu-
(and hence a nonnegative quantity) of the ith subsystem, cidate that (3) is essentially the statement of the principle of
let si : R+ + B denote the external power (heat flux) the conservation of energy let the total energy in the large- -
supplied to (or extracted from) the ith subsystem, let u;j : scale dynamical system B be given by U c eTVs, V, E , :
W
--v where eT 11, ...,l],and let the energy received by
Et+ iE+, i # j , i,j = 1,._.,q. denote the instantaneous
rate of energy (heat) flow from the jth subsystem to the ith the large-scale dynamical s stem D over the time interval
subsystem, and let U;; : n:-+E' i = 1,...,q. denote the [tl,tz] he given by Q k !J eT[S(t) - d(V.(t))]dt,where
K ( t ) ,t 2 t o , is the solution to (6). Then, premultiplying
instantaneous rate of energy (heaa'dissipation from the ith
subsystem to the environment. Hence, an energV balance (3) by eT and using the fact that eTw(VJ = 0, it follows
equation for the ith subsystem yields that
v.i(T) = %;(to) AU = Q, (7)

+ 2 / T [ U i j ( K ( t ) ) - ~ji(K(t))ldt
3 = l , j # i 60
where AU 4 U(t2) - U(t1) denotes the variation in
energy of the large-scale dynamical system D over the
time interval [ t l , t z ] . This is a statement of the first law

/ T of thermodynamics for the large-scale dynamical system D

-lQTui;(K(t))dt+ t o si(t)dt, T t t o , (2) and gives a precise formulation of the equivalence between
variation in system internal energy and heat. It is important
to note that our large-scale dynamical system model does
or, equivalently, in vector form, not consider work done by the system on the environment

/ T
K V ) = K(to) + w(l/s(t))dt - d(K(t))dt that
to
4 T nor work done by the environment on the system. Hence,
Q can be interpreted physically as the amount of energy
is received by the system in forms other than work.

+ loT S(t)dt, T 2 t o ,
The extension of addressing work performed by and on the
(3) system can be easily handeled by including an additional
state equation, coupled to the power balance equation (6),
involving volume states for each subsystem with exogenous
where K(t) [v.~(t),....u.qa(t)]T, d ( K ( t ) ) C pressure variables. Since this slight extension does not alter

t 2 t o , and zu = [wl,...,w,]T -
bIl(K(t)), ..., u g q ( K ( t ) ) l TS. ( t ) = [ S l ( t ) , ..., sp(t)lT, any of the results of the paper, it is not considered here for
: E t Rq is such that simplicity of exposition..
If the total energy of the large-scale dynamical system
D at the initial and the final states is fixed, then it follows
from (7) that the variation (6) of the energy supplied to the
large-scale dynamical system D is zero; that is, SQ = 0.
399
 This implies that during a transformation between two fixed potential V,(-T) = 0 to a given state V.(to) = V,,, it
end points the large-scale dynamical system G receives a follows from (9) that the large-scale dynamical system G
fixed amount of energy. In other words, for any two paths can deliver to its surroundings all of its stored subsystem
connecting the initial and final states of the dynamical energies and can store all of the work done to all of its
system G the amount of energy supplied to the system is subsystems. In the case where S ( t ) 0 it follows from (8)
the same. and the fact that uii(V.) 2 0, V , E E:, i = 1,...,q , that
If U,(&) = 0, V , E -R,, ‘I
whenever vsj = 0, i,j = the zero solution V,(t) 0 of the large-scale dynamical
a:,
1,. . . , q , then w(V.) - d ( V , ) , V, E is essentially non- system G with the power balance equation (6) is Lyapunov
stable with Lyapunov function U ( % ) corresponding to the
negative. The above constraint implies that if the energy of
the jth subsystem of G is zero, then this subsystem cannot total energy in the system.
supply any energy to its surroundings nor dissipate energy to The nonlinear power balance equation (6) can exhibit a
the environment. Moreover, for the remainder of the paper full range of nonlinear behavior including bifurcations, limit
we assume that si(t) 2 0 whenever u,i(t) = 0, t 2 to, cycles, and even chaos. However, a thermodynamically con-
i = 1,..., q, which implies that when the energy of the ith sistent energy flow model should ensure that the evolution
subsystem is zero, then no energy can be extracted from of the system energy is diffusive (parabolic) in character
this subsystem. The following proposition is needed for the with convergent subsystem energies. Hence, to ensure a
main results of this paper. thermodynamically consistent energy flow model we require
Proposition 3. I : Consider the large-scale dynamical sys- the following axioms. For the statement of these axioms we
tem G with power balance equation given by (6). Suppose first recall the following graph theoretic notions.
-P
uij(V,) = 0, V , E R+, whenever usj = 0, i , j = 1,...,q, Definition 3.1 ([37]): A directed graph G(C) associ-
and s,(t) 2 0 whenever w.i(t) = 0, t 2 to, i = 1,..., q. ated with the connectivity matrix C E RqXq has vertices
Then the solution V.(t),t 2 t o ,to (6) is nonnegative for {l,2 , ...,q } and an arc from vertex i to vertex j, i # j,
all nonnegative initial conditions E E:. if and only if C ( j , i ) # 0. A graph G(C) associated with
the connectivity matrix C E RqXqis a directed graph for
Proof. First note that w(V,) - d ( V , ) , V , E @+, is which the arc set is symmetric; that is, C = CT. We say
essentially nonnegative. Next, since si(t) 2 0 whenever that G(C), is strong? connected if for any ordered pair of
v.i(t) = 0, t 2 to, i = 1,...,q, it follows that &(t) 2 0 vertices z , j ) , z # 3. there exists apath (i.e., sequence of
for all t 2 t o and i = 1, . . . , q whenever v,i(t) = 0 and arcs) leading from i to j.
u.j(t) 2 0 for all j # i and t 2 to. This implies that for all
Recall that C E R q x q is irreducible; that is, there does
nonnegative initial conditions V . 0 E $+ the trajectory of G not exist a permutation matrix such that C is cogredient
is directed towards the interior of the nonnegative orthant to a lower block triangular matrix, if and only if G(C)
Y
R, whenever vSi(t) = 0, i = 1,...,q, and hence remains is strongly connected (see Theorem 2.7 of [37]). Let
nonnegative for all t 2 t o . U &j(V,) u;j(V+uji(K), V , E , :
E definethenet energy
flow from the jth subsystem G, to the ith subsystem Gi
- :
Next, premultiplying 3) by eT, using Proposition 3.1,
and using the fact that e$.I(&) = 0, it follows that
of the large-scale dynamical system G. Axiom i): For the
connectivity matrix C E RqXq associated with the large-
scale dynamical system S defined by
eTV,(T) = eTV,(to) + /to
T
eTS(t)dt
{ y: if $ij(V,) = 0, .
otherwise, z # j , i , j = l ,... q,(12)
(

- ioTeTd(K(t))dt, T 2 to. (8) and

Now, for the large-scale dynamical system G define the

input u(t) S ( t ) and the output y ( t ) e d ( V . ( t ) ) .Hence, it
C(. . - -
*>%) f:
k=l, k j i
C(k.0, i=j, i = l , ...,q, (13)
follows from (8) that the large-scale dynamical system G is
lossless [27] with respect to the supply rate eTu - eTy and rank C = q - 1 and for C ( i , j ) = 1, i # j , $,j(V,) = 0
with the storage function U ( V , ) eTV,, V , E . :
E This if and only if uSi = v8j. Axiom ii): For i,j = 1,...,q,
implies that (see [27] for details) (u.i-vsj)&,(Vs) 2 0, V, €E:. The fact that &j(V,) = 0
if and only if vsi = vsj, 2 # j , implies that subsystems
0 5 U&) = U ( & ) = U&) < 00, v, E E
!, (9) E, and G j of G are connected; altematively, &(V,) = 0
where implies that Gi and G j are disconnected. Axiom i) implies
that if the energies in the connected subsystems G< and Gj
are equal, then energy exchange between these subsystems
is not possible. This statement is consistent with the zeroth
law of fhermoQnamics which postulates that temperature
equality is a necessary and sufficient condition for thermal
e uilibrium. Furthermore, it follows from the fact that C =
C9. and rank C = q - 1 that the connectivity matrix C is
Since U,(V,) is the maximum amount of stored energy irreducible which implies that for any pair of subsystems
which can be extracted from the large-scale dynamical Gi and Gj, i # j, of G there exists a sequence of connectors
system G at any time T and U,(&) is the minimum (arcs) of G that connect G, and Gj: Axiom ii) implies
amount of energy which can be delivered to the large-scale that energy flows from more energetic subsystems to less
dynamical system Q to transfer it fiom a state of minimum energetic subsystems and is consistent with the second law
400
of thermoc+namics which states that heat (energy) must flow
in the direction of lower temperatures. Furthermore, note
that q&(V,) = -&(Vs), V, E E:, i # j , i , j = 1, ..., q,
which implies conservation of energy between lossless
subsystems. With S ( t ) 0, Axioms i) afld ii) along with
the fact that &(Vi) = -$ji(V,), V. E R I , i # j,i , j =
1, .._, q, imply that at a given instant of time energy can only
be transported, stored, or dissipated but not created and the
maximum amount of energy that can he transported and/or
dissipated from a subsystem cannot exceed the energy in
the subsystem.
Next, we establish that the classical Clausius inequality
for reversible and irreversible thermodynamics is satisfied 5 0,
for our thermodynamically consistent energy flow model.
For the remainder of the paper we assume that the large- which proves the result.
scale dynamical system B with the power balance equation
( 6 ) is reachable from and confrollable to the origin in E!. Inequality (14) is Clausius' inequality for reversible and
Recall that the large-scale dynamical system D with the irreversible thermodynamics as applied to large-scale dy-
power balance equation ( 6 ) is reachable from the origin in namical systems. It follows from Axiom i) and ( 6 ) that
-q
R, if, . for all V,O = V(t0)E El,there exists a finite for the isolated large-scale dynamical system D; that is,
time ti 5 t o and a square integrable input S ( t ) defined on S ( t ) 0 and d(V'(t)) = 0, the energy states given by
[ti,t o ] such that the state K ( t ) ,t 2 ti, can be driven from K, = (Ye; cy 2 0, correspond to the equilibrium energy
V,(ti)= 0 to %(to) = KO.Alternatively, D is controllable states of 9. Thus, as in classical thermodynamics, we can
to the origin in E, if, for all Vh = K(t0) E E ' there define a reversibleprocess as a process where the trajectory
exists a finite time tr 2 t o and a square integrabli input of the large-scale dynamical system G moves along the
S ( t ) defined on [to,tr] such that the state K ( t ) ,t 2 to, set of equilibria of the isolated system D. The power
can be driven from V.(to) = Vap to V,(tr) = 0. We let in ut that can enerate such a trajectory can be given by
U, W ' denote the set of all admlssible power inputs (heat S b ) = d(K(t)! + u(t),t 2 to, where U ( . ) E Rq is such
flux) to the large-scale dynamical system D such that for that ui(t) u j ( t ) ,i # j , i , j = 1, ..., q. Alternatively,
any T 2 -to the system energy-state can be driven from an irreversible process is a process that is not reversible.
Hence, it follows from Axiom i) that for a reversible process
K(-7') = 0 to K(t0) = KOE by S(.) E U, and we
& ( K ( t ) ) 0, t 2 to, i # j,i , j = 1,._.,q, and thus, by
let U, & W ' denote the set of all admissible power inputs Proposition 3.2, inequality (14) is satisfied as an equality.
(heat flux) to the large-scale dynamical system B such that Alternatively, for an irreversible process it follows from
for any T 2 t o the - system energy state can be driven from
Axioms i) and ii) that (14) is satisfied as a strict inequality.
V,(to) = KO E Wt to V,(T) = 0 by S(.)E U,.For the
next result f denotes a cyclic integral evaluated along an Next, we give a deterministic definition of entropy for
the large-scale dynamical system D that is consistent with
arbitrary closed path of (6) in $+,that is, f +? with the classical thermodynamic definition of entropy.
t r 2 t o and S(.) E W q such that V,(tr)= V,(t,) = KOE
-4 Definition 3.2: For the large-scale dynamical system D
W+. with the power balance equation ( 6 ) , a function S : E: +
Proposition 3.2: Consider the large-scale dynamical sys- R satisfying
tem G with the power balance equation (6) and assume that
Axiom ii) holds. Then for all V,O E , :
E tr 2 t o , and S ( V ( t 2 ) ) 2 S(K(t1))
S ( t ) ,t E [to,tr], such that &(it) = KO,

for any t z 2 t i 2 t o and S ( t ) , t E [ t l , t 2 ] , is called the

entropy of 9.
Next, we show that (14) guarantees the existence of an
where c > 0, d&i(t) [ s ; ( t ) - u;;(V,(t))]dt, i = 1, ...,q, entropy function for 9.For this result define
is the amount of energy received by the ith subsystem over
the infinitesimal time interval dt, and K ( t ) ,t 2 t o , is the
solution to (6) with initial condition K ( t o )= KO.
Proof. Since, by Proposition 3.1, V,(t) 22 0, t 2 to,
and &(K) = -+j;(K), V. E , :E i # j , i , j = 1, ..., q. it
follows from ( 6 ) and Axiom ii) that

401
 Theorem 3.1: Consider the large-scale dynamical system Note that with V,(to) = V,(T)= 0 it follows from (14)
G with the power balance equation (6) and assume that that supremum in (17) is taken over the set of nonpositive
Axiom ii) holds. Then there exists an entropy function for values with one of the values being zero for S ( t ) 0. Thus,
G. Moreover, Sa(V,), V , E E:, and &(Vi),V , E Wt,are S,(O) = 0. Similarly, it can be shown that S,(K), V , E E:,
possible entropy functions for G with Sa@) = SJO) = 0. given by (18) satisfies (16) and hence is a possible entropy
Finally, all entropy functions S(V,), V , E a:, for G satisfy function for the system 0 with SJO) = 0.
Next, suppose there exists an entropy function S : +
Sr(V,) I S(K) - S(0) I Sa(K), V , E E:. (19) W for G and let V , ( t 2 ) = 0 in (16). Then it follows from
Proof. Since G is controllable to and reachable from the (16) that
origin in E : it follows from (17) and (18) that S,(V,o) <
- Y
03, V,OE , W and S,(Vs0)> -m, V.0 E R,, respectively.
:
Next, let V,O E E?+and let S ( t ) ,t E [ti,tf]. where ti I
to .S tf, be such that &(ti) = V,(tf) = 0 and V,(to)=
In this case, it follows from (14) that

or, equivalently,

Now, taking the supremum on both sides of (21) over all

S(.) E U,and ti I to yields

Next, taking the infimum on both sides of (22) over all

= S,(V,(tZ)), (28)
S E U, and tf 2 to we obtain Sr(K0)I Sa(!&), KOE
-(.)
W,, which implies that -03 < S,(K0) 5 &(KO) < which, since V,(tz) is arbitrary, implies that S,(V,) I
03, KO E E :
. Hence, the functions Sa(.)and S,(.) are S(V.) - S(O), V. E 9,.Thus, all entropy functions for
well defined. Next, it follows from the definition of S a , 4 satisfy (19). 0
that, for any T 2 tl and S ( . ) E U, such that & ( t l )E W,
and K ( T ) = 0, Remark 3.1: It is important to note that inequality (14)
is equivalent to the existence of an entropy function for
9. Sufficiency is simply a statement of Theorem 3.1 while
necessity follows from (16) with V,(t2) = V,(tl). For
irreversible thermodynamics with power balance equation
(6), Definition 3.2 does not provide enough information to
define the entropy uniquely. This difficulty has long been
pointed out in [39]. For reversible thermodynamics this
ambiguity is not an issue as (14) holds as an equality for
and hence a reversible process since &(V.(t)) 0, i # j , i , j =
1,...,q, and in this case it can be shown that S,(V,) =
S,(K) = s(4) - S(O)= eTlog,(ce + Vi) - qlog,c,
where V , = V , , and log,(ce + V,) denotes the vector
natural logarithm given by [log,(c+v,l), ..., log,(c+~.,)]~.
This definition of entropy leads to the second law of
thermodynamics being viewed as an axiom in the context
of (anti)cyclo-dissipative dynamical systems [27], [40]. A
similar remark holds for the definition of ecisopy mtroduced
below.
The next result shows that all entropy functions for G are
which implies that S,(V,), V , E E:, satisfies (16). Thus, continuous on E:. First, however, the following lemma is
S,(V,), V, E E l , is a possible entropy function for G. required.
402
 Lemma 3.1: Consider the large-scale dynamical system S : (0,f?] -+ Hpq such that IlS(t) - Sell < E , t E (0,f), and
G with power balance equation (6). Then for every equi- E(2) = Eo. In addition, it follows from Lemma-3.1 that
librium state E, E E: and every E > 0 and T > 0, S : (O,?] -t Rq is such that E ( t ) 2 1 0, t E [O,T].Next,
there exists Se E Rq, (Y > 0, and T E [O,T]such that since u i i ( . )i, = 1 , .. . ,q, is continuous it follows that there
for every E E E: with Ilk - E,[[ I (YT, there exists exists M E (0, CO) such that
S : [O, T ]+ Rq. such that llS(t)- Sell 5 E , t E (0, 'f], and
E ( t )= E, + Wt,t E (O,?]. SUP
llE-Ee11<6, /IS--Sel/<r i = l
= M. (32)
Proof. Note that with Se = d(E,) - w(E,),the state
E, E zg"+ is an equilibrium state of (6). Let 0 > 0 and Hence, it follows that
T > 0 and define

+Sell. (29)
Note that for every T > 0, lime-o+ M ( 0 , T ) = 0 and for
every 8 > 0, limT,o+ M ( 0 , T ) = 0. Next, let E > 0 and
T > 0 be given and let OL > 0 be such that M ( a ,T)+a 5 E . 5 Mf 5 M T 5 y-'([[Eo- Eel[). (33)
(Note that a 5 E; the existence of such an p is guaranteed Now, if S(.) is an entropy function of 9, then
since M ( u , T ) + 0 as CY -t Ot.) Now, let E E a: be such
that 113 - E1.1 5 aT. With 'f 4 5 T and

or, equivalently,

it follows that

If S(E,) 2 S(E(q),then combining (33) and (35) yields

is a solution t? ( 6 ) . The result is now immediate by noting
that E ( T ) = E and IS(&) -S(E(O)I Iy-'(lIEo - &/I). (36)
Altematively, if S(E(f))2 S(E,), then (36) can be derived
by reversing the roles of E and E(t^).In particular, using
the fact that G is locally controllable from and to E,,
similar arguments can be used to show that the set of
points which can hesteered in small time to E, contains a
+
5 M ( a , T ) a 5 E, t E [ O , T ] . (31) neighborhood of E ( t ) .Hence, since y(.) is continuous and
--P
U E(2) is arbitrary, it follows that S(.) is continuous on R,.
U
Theorem 3.2: Consider the large-scale dynamical system
D with the power balance equation (6) and let S : i R The next proposition shows that if (16) holds as an
be
- an entropy function of 0. Then S ( . ) is continuous on equality for some transformation starting and ending at
"'+ . an equilibrium point of the isolated system G, then this
transformation must be reversible.
Proof. Let E, E a: and Se E BY be such that set =
Proposifion 3.3: Consider the large-scale dynamical sys-
U,%(&) - C:=l,lft&(E,), i = 1 , . . .,q. Note that with tem G with the power balance equation (6) and assume that
S(1) = Se, E, is an equilibrium point of the power balance Axioms i) and ii) hold. Let S(.) denote an entropy of G
equation ( 6 ) . Next, it follows from Lemma 3.1 that 0 is and let E : [to, t l ] + denote the solution to (6) with
locally controllable; that is, for every T > 0 and E > 0 E(to)= aoe and E ( t l )= Lyle, where eo, cyI 2 0. Then,
the set of points which can be reached from and to E, in
time T using admissible inputs S : [0,T ]--t Rq,satisfying
S ( t ) -Sell < E, contains a neighborhood of E,. Next, let
> 0 and note that it follows from the continuity of w(.)
and d(.) that there exist T > 0 and E > 0 such that for every
+ Rq and IS(t) Se 1 iE , [IE(t)- E,[! < 6, if and only if there exists a : to,tl] -t E+ such that
where E(t\, 1 E TO, T\, denotes the solution to L y ( t o ) = eo,a ( t l )= el, and E ( t \ = a(t)e,t E [to,tl].
(6) with the initial condition E,. Furthermore, it follows
Proof. Since E(to)and E(t1) are equilibrium states of
from the local controllability of G that for every T E (0, TI the isolated system G it follows from Remark 3.1 that
there exists a strictly increasing, continuous function y :
€3t R such that $0) = 0 and for every Eo E %: such S(E(t1))- S(E(t0)) = 4log,(c + a i )
that 1lE0- E.\\ 5 y(?), there exists t^ E (a,?] and an input -QlOg,(C+ao). (38)
403
Furthermore, it follows from (6) that The entropy expression given by (40) is identical in form
to the Boltzmann entropy for statistical thermodynamics.
Due to the fact that the entropy is indeterminate to the
extent of an additive constant, we can place the constant
of integration qlog, c to zero by taking c = 1. Since S(V,)
given by (40) achieves a maximum when all the subsystem
energies U,;, i = 1,...,q. are equal, entropy can be thought
of as a measure of the tendency of a system to lose the
ability to do useful work and lose order and to settle to a
more homogenous state.
Recalling that dQ;(t) = [s;(t)- u;;(Vs(t))]dt, i =
1,...,q, is the infinitesimal amount of heat received or
dissipated by the ith subsystem of G over the infinitesimal
time interval dt, it follows from (16) that
Now, it follows from Axioms i) and ii) that (37) holds if
and only if e ; ( t ) = e j ( t ) , t E [to,tll, i # j , i , j = 1,...,q,
or equivalently, there- exists a : [to,tl] -+ E+ such that
E ( t )= a(t)e,t E [to,tl].
Inequality (42) is inslogous to the classical thermodynx"
The next proposition gives a closed-form expression for inequality for the tion on o i entropy during an infinitesi-
the entropy of G. mal incversihle transformation with thc shitted subsystem
cnergiss c vs, playing the role of the ith subsystem
Proposition 3.4: Consider the large-scale dynamical sys- thermodynamic (absolute) tcmperaturcs.
tem G with the power balance equation (6) and assume that
Axiom ii) holds. Then the function S : E! E% given by h'cxr, we introduce 3 tieit' 2nd dual notian IO entropy;
namcly ectropy, dcscribing thc status quo of the liirgc-
+
S(Va) = eTlog,(ce V.) - qlog,c, V, E E:, (40) scale dynamical system G. First, howsver, W O prescnt a
dual to inequality (141 that holds for our thermodynamically
where c > 0, is an entropy function of 6. consistent energy now model.
Proof. Since, by Proposition 3.1, V.(t) 22 0, t 2 to, Pn,poJiriun 3 5: Consider the large-scale dynamical sys-
and4ij(V,)=-4jji(V.) , V . E ~ + , i f j , i , j = l , . . . , q,it tem G with the powur balsnce quation ( 6 ) and ilssumr that
follows that Axiom 7 1 ) holds. Then for all l'sd E a:.
I , 2 t u . and
.S(t). r c [ t , , , t , ] , such that V,(ttj = \ib,

where K ( t ) , t 2 to, is the solution to (6) with initial

condition V,(to)= Vso.
Proof. Since, by Proposition 3.1, & ( t ) 22 0, t 2 to,
andOij(V,)=-4ji(l/,) , V . E I W ~ , i f j , i , j = l , . . . ,q,it
follows from (6) and Axiom i i ) that

Now, integrating (41) over Itl, t z ] yields (16). 0

Remark 3.2: Note that it follows from the last equality in
(41) that the entropy function given by (40) satisfies (16) as 2 0, (44)
an equality for a reversible process and as a strict inequality
for an irreversible process. which proves the result. 0
404
 Note that inequality (43) is satisfied as an equality large-scale dynamical systems; while inequality (45) is an
for a reversible process and as a strict inequality for an anti Clausius inequality. Moreover, for the ectropy function
irreversible process. Next, we present definition of ectropy defined by (49), it follows from Proposition 3.6 that a
for the large-scale dynamical system 9. thermodynamically consistent large-scale dynamical system
Definition 3.3: For the large-scale dynamical system 9 model is dissipative with respect to the supply rate KTS and
with storage function corresponding to the system ectropy.
with the power balance equation (6), a function E : --t R
For the entropy function given by (40) note that S(0) = 0
satisfying which is consistent with the third law of thermodynamics
E(K(t2)) 5 L(V,(td) (Nemst’s theorem) which states that the entropy of every
system at absolute zero can always be taken to be equal to
zero. For the isolated large-scale dynamical system G, (16)
yields the fundamental (universal) inequality

“(tz)) 2 S(V.(tl)), t2 2 tl. (50)

Inequality (50) implies that, for any dynamical change in
For the next result define an isolated large-scale dynamical system 0,the entropy of
the final state can never be less than the entropy of the
initial state. It is important to stress that this result holds for
an isolated dynamical system. It is however possible with
power (heat flux) supplied from an external system to reduce
the entropy of the dynamical system 9. The entropy of
-q
where V,(t,)= V,, E R, and V,(T) = 0, and both systems taken together however cannot decrease. The
above observations imply that when the isolated large-scale
&SV.O) dynamical system 9 with thermodynamically consistent
energy flow characteristics (i.e., Axioms i) and ii) hold)
is at a state of maximum entropy consistent with its energy,
it cannot be subject to any further dynamical change since
any such change would result in a decrease of entropy. This
where V,(-T) = 0 and V,(to)= V,O E E : of course implies that the state of maximum entropy is the
stable state of an isolated system and this equilibrium state
Theorem 3.3: Consider the large-scale dynamical system has to be semistable. Analogously, it follows from (45) that
G with the power balance equation (6) and assume that the isolated large-scale dynamical system g satisfies the
Axiom ii) holds. Then there exists an ectropy function for fundamental inequality
9. Moreover, E,(&), V , E Et, and E,(%), V , E E!, are
possible ectropy functions for 9 with &a(0) = E,(O) = 0. “2)) 5 E ( v , ( t l ) ) , t z 2 tl, (51)
Finally, all ectropy functions E(V,),V , E F+,
for 0 satisfy
which implies that the ectropy of the final state of 0 is
Ea(&) V, E p+, (48) always less than the ectropy of the initial state of 0.Hence,
5 &(K)- E ( 0 ) 5 E r ( % ) ,
for the isolated large-scale dynamical system B the entropy
Proof. The proof is similar to the proof of Theorem 3.1. increases if and only if the ectropy decreases. Thus, the
0 state of minimum ectropy is the stable state of an isolated
system and this equilibrium state has to be semistable. The
The next proposition gives a closed-form expression for next theorem concretizes the above observations.
the ectropy of 9. Theorem 3.4: Consider the large-scale dynamical system
Proposition 3.6: Consider the large-scale dynamical sys- 0 with power balance equation (6) with S t) 0 and
tem D with the power balance equation (6) and assume that d ( K ) E 0 and assume that Axioms i) and i!) h z d . Then
Axiom ii) holds. Then the function E : Et + P given by for every a 2 0, a e is a semistable equilibrium state of
(6). Furthermore, V.(t) + i e e T V , ( t o ) as t + 00 and
&(V,)= ;v:vs, v, E Et? (49) iV e e T l < ( t o ) is a semistable equilibrium state. Finally, if
is an ectropy function of 9. for some k E {l, ...,q } , o k k ( K ) 2 0, V, E E;“+, and
ukk(Vs) = 0 if and only if 2i.k = Oz, then the zero
Proof. The proof is similar to the proof of Proposition solution V,(t)= 0 to (6) is a globally asymptotically stable
3.4. 0 equilibrium state of (6).
Remark 3.3: Note that the ectropy function given by (49) Proof. It follows from Axiom i) that a e E a:,
a 2 0, is
satisfies (45) as an equality for a reversible process and as an equilibrium state for (6). To show Lyapunov stability of
a strict inequality for an irreversible process. the equilibrium state a e consider the system shifted ectropy
function E,(K) = ;(K - a e ) T ( V , - a e ) as a Lyapunov
It follows from (49) that ectropy is a measure of the function candidate. Now, since &(V8) = -&(VS), V, E
extent to which the system energy deviates from a homo-
geneous state. Thus, ectropy is the dual of entropy and is a
measure of the tendency of the large-scale dynamical system 2The assumption ukk(V.) 2 0 , V. E , :
E and ukk(V,) = 0 if and
L7 to do useful work and grow more organized. only if v.* = 0 for some k E {I, ...,q } implies that if the kth subsystem
possesses no energy, then this subsystem cannot dissipate energy ta the
Inequality (16) is precisely Clausius’ inequality for re- environment. Conversely, if the kth subsystem does not dissipate energy
versible and irreversible thermodynamics as applied to to the environment, then this subsystem has no energy.

405
E:, i # j , i , j = 1,...,q, and eTw(V,)= 0, V , E E:, it Lyapunov stable. To show global asymptotic stability of the
follows from Axiom ii) that zero equilibrium
- state let R 6 {Vs E p +: &(V,)= 0 ) =
{Vs E : u S k u k k ( V . ) = 0, k E 11,. . . , q } } n {V, E
&(K) = (v,-ae)Tri, -
a',: (us; - us.)&(Vs) = 0 , i = 1,...,q , j E IC;}. Now,
= (V, - ae)Tw(V,) since Axiom ij holds and u k k ( V , ) - = 0 if and only if
~ 0 it follows that R = {V, E Wt : usk = 0 , k E
I J ,=
{I,...,q } } n {v,E R', : vsl = u , ~= . . . = w s9 } = {o}
and the largest invariant set M contained in R is given
by M = {O}. Hence, it follows from the Krasovskii-
LaSalle invariant set theorem that for any initial condition
V,(to)E Et, K ( t ) + M = {0} as t + m which proves
global asymptotic stability of the zero equilibrium state of
(6). 0

In Theorem 3.4 we used the shifted ectropy function to

show that for the isolated ( i s . , S ( t ) E 0 and d(V,) z 0)
5 0, V , ER',, (52) large-scale dynamical system G, V.(t) + :eeTV,(to) as
ti 05 and $eeTV,(to) is a semistable equilibrium state.
where X i 5 N,\ U:zi{l} and N i 5 { j E {l, ...,q} : This result can also be arrived at using the system entropy.
&(V,) = 0 if and only if U," = u s j } , i = 1,..., q , which To see this note that since eTw(V,) = 0, V , E R!, it
establishes Lyapunov stability of the equilibrium state ae.
To show that ae is semistable, let R 5 {V, E R: : follows that eTV3(t) = 0, t 2 to. Hence, eTV,(t) =
eTV,((to),t 2 to. Furthermore, since V,(t)22 0, t 2 to,
€&) = 0 ) = {V, E E: : (V,i - ? J s j ) $ l i j ( V , ) = 0 , i =
it follows that 0 55 V,(t)55 eeTV,(t,),t _> to, which
1,..,,q, j E X i } . Now, by Axiom i) the directed graph implies that all solutions to (6) are bounded. Next, since by
associated with the connectivity matrix C for the large-scale (50) the entropy S(G(t)), t 2 to, of G is monotonically
dynamical system G- is strongly connected which implies increasing and V , ( t ) ,t 2 t o , is bounded, the result follows
that R = { V ~E w', : uS1 = . . - us,}. Since the using similar arguments as in Theorem 3.4.
set R consists of the equilibrium states of (6), it follows
that the largest invariant set M contained in R is given by Theorem 3.4 implies that the steady-state value of the
M = R.Hence, it follows from the Krasovskii-LaSalle energy in each subsystem 4; of the isolated large-scale
invariant set theorem 1411 that for any initial condition dynamical system 4 is equal; that is, the steady-state energy
of the isolated large-scale dynamical system G given by
V,(to) E E',, V , ( t ) , i M as t + m and hence cue
is a semistable equilibrium state of (6). Next, note that
since eTV,(t) = eTV,(to) and K ( t ) + M as t + 05,
it follows that V s ( t ) i ieeTV,(to) as t + 05. Hence,
I
v,, = +eeTK(to) = [ ~ ~ ~ = i u . , ( t eo )is uniformly
distributed over all subsystems of 9. T is phenomenon
is known as equipartition of energy' [23], [261, [291,
[42], [43] and is an emergent behavior in thermodynamic
with a = :eTV,(to), ae = ieeTV,(to) is a semistable
equilibrium state of (6). systems. The next proposition shows that among all possible
energy distributions in the large-scale dynamical system 8,
Finally, to show that in the case where for some k E energy equipartition corresponds to the minimum value of
{I, ...,q}, u k k ( V , ) 2 0, V , E E',,and akk(V,) = 0 if and the system's ectropy and the maximum value of the system's
only if v,k = 0, the zero solution V,(t)E 0 to (6) is globally entropy (see Figure 2).
asymptotically stable, consider the system ectropy E ( & ) = Proposition 3.7: Consider the large-scale dynamical sys-
$KTV,, V , E Et, as a candidate Lyapunov function. Note tem G with power balance equation (6), let E : g+.-t L+
that &(O) = 0, &(V,)> 0 , V, E ?+, V, # 0, and &(V,) and S : E', + W denote the ectropy and entropy of G given
is radially unbounded. Now, the Lyapunov derivative along
the system energy trajectories of (6) is given by by (49) and (40), respectively, and define V, {V. E E ! :
eTV, = b}, where b 2 0. Then,
t(W = v:rw(v%) - d(K)1
P
ar min & V , ar max(S(V,)) = V: = -e.
$EDc ( ( ) ) = E,, 9
(54)

Furthermore, Emi, i+ & ( y=);$ and S

,, d S(V,') =
qlog,(c+ $) - qlog,c.
Proof. The existence and uni ueness of V: follows
from the fact that E(V,) and -SlV,) are strictly convex
continuous functions defined on the compact set Vc.To
minimize &(V,) = f V z V , , V, E E',, subject to V. E V c
form the Lagrangian L(K,X) = :KTK +
X(eTK - P),
where X E P is the Lagrange multiplier. If V,' solves this

'The phenomenon of equipanition of energy is closely related l o C c

notion of a monoremperoruric system discussed in 1251.

406
 dynamical system G all the energy, though always con-
served, will eventually be degraded (diluted) to the point
where it cannot produce any useful work. Hence, all motion
would cease and the large-scale dynamical system would
be fated to a state of etemal rest (semistability) wherein a!l
subsystems will posses identical energies (energy equiparti-
tion). Ectropy would be a minimum and entropy would be
a maximum giving rise to a state of absolute disorder. This
is precisely what is known in theoretical physics as the heut
death of the universe.
Next, we show that our thermodynamically consistent
large-scale system G satisfies Gibbs'principle [U,p. 561.
Gibhs' version of the second law of thermodynamics can
be stated as follows:
Gibbs' Principle. For an equilibrium of any isolated
system it is necessary and suficient that in all possible
variations of the state of the system which do not
" alter its e n e w , the variation of its entropy shall either
el
vanish or be negative.
Fig. 2. Thermodynamic equilibria (. .), constant energy
To establish Gihb's principle for our thermodynamically
surfaces (- ), constant ectropy surfaces (- - -), and consistent energy flow model, suppose E, = a e , a 2 0,
constant entropy surfaces (- . - . -) is an equilibrium point for the isolated system 0. Now, it
follows from Proposition 3.7 that the entropy of G achieves
its maximum at E. subject to the constant energy level
minimization problem, then eTE = aq, E E E:. Hence, any variation of the state
of the system which does not alter its energy leads to a
= v:~ + XeT = o (55)
zero or negative variation of the system entropy. Conversely,
suppose at some point E' E E,! the variation of the system
entropy is either zero or negative for all possible variations
and hence V: = -Xe. Now, it follows from eTV, = fi in the state of the system which do not alter the system's
that X = - a which implies that V,. = f e E E:. The fact total energy. Furthermore, ad absurdum, let the isolated
that V: mi2mizes the ectropy on the compact set D, can system G undergo an irreversible transformation starting at
be shown by computing the Hessian of the ectropy for the E'. Then it follows from Proposition 3.4 that the entropy
constrained parameter optimization problem and showing of G given by (40) strictly increases which contradicts the
above assumption. Hence, the system G cannot undergo an
that the Hessian is positive definite at V,..&jn = 4$ is irreversible transformation starting at E'. Altematively, if
now immediate. the isolated system 0 undergoes a reversible transformation
Analogously, to maximize S(V,) = eTlog,(ce V , ) - + starting at E', then E* has to be an equilibrium state of 8.
qlog,c on the compact set Vc,form the Lagrangian In the preceding discussion it was assumed that our large-
L(V,,X) 4 ~ ~ = l l o g , ( c + u , i ) + X ( e T V s - f i ) , where X E E
4 scale dynamical system model is such that energy flows
is a Largange multiplier. If V,' solves this maximization from more energetic subsystems to less energetic subsys-
problem, then tems; that is, heat (energy) flows in the direction of lower
temperatures. Although this universal phenomenon can be
predicted with vimal certainty, it follows as a manifestation
of entropy and ectropy nonconservation for the case of two
1 subsystems. To see this, consider the isolated large-scale
dynamical system G with power balance equation (6) (with
c +u t S(t) 0 and d ( K ) = 0) and assume that the system
entropy given by (40) is monotonically increasing and hence
Thus, X = --, i = 1,..., q. If X = 0, then the only S(V,(t))2 0, t 2 to. Now, since
value of V,.that satisfies (56) is V t = 00, which does not
satisfy the constraint equation eTV, = p for finite fi 2 0.
Hence, X # 0 and U:; = - ( Lx + e ) , i = 1, ..., q, which
implies V: = -($ +
c)e. Now, it follows from eT1/, = 0
that -(I + e ) = e9 and hence V: = :e E E:. The fact
that Vc maximizes the entropy on the compact set V, can
be shown by computing the Hessian and showing that it is
negative definite at v.
,,S
, = qlog,(c + 5)
- qlog, c is
now immediate. 0

It follows from (50), (51), and Proposition 3.7 that

conservation of energy necessarily implies nonconservation
of ectropy and entropy. Hence, in an isolated large-scale
 2 0, t 2 to, (57) hence the system entropy S(V,)is an additive quantity over
all subsystems. i w ) For the system 9, S(V,) 2 0 for all
) 0, V , E E:,
it follows that for q = 2, (wsl - w , z ) ~ l z ( ~ , 5 V, E w) It follows from Proposition 3.7 that for a
which implies that energy (heat) flows naturally from a given vatue p 2 0 of the total energy of the system G,
more energetic subsystem (hot object) to a less energetic one and only one state; namely, V,. = $e, corresponds
subsystem (cooler object). The universality of this emer-
gent behavior thus follows from the fact that entropy (re- to the largest value of S(V,). vi) It follows from (40)
spectively, ectropy) transfer, accompanying energy transfer, that for the system 8 , gra h of entropy versus energy
always increases (respectively, decreases). In the case where is concave and smooth. w i i f For a composite large-scale
we have multiple subsystems, it is clear from (57) that dynamical system Gc of two dynamical systems GA and GB
entropy and ectropy nonconservation does not necessarily the expression for the composite entropy Sc = SAt SB,
imply Axiom ii). However, if we invoke the additional where SA and SBare entropies of GA and GB, respectively,
condition (Axiom iii)) that if for any pair of connected is such that the expression for the equilibrium state where
subsystems Gk and G[, k # I , with w.k 2 w.1 (respectively, the composite maximum entropy is achieved is identical to
w.k 5 w.1) and for any other pair of connected subsystems those obtained for GA and GB individually. Specifically, if
Gm and G,,, m # n, with w, 2 w,, (respective1 w, 5 pA and qB denote the number of subsystems in GA and GB,
respectively, and PA and PB denote the total energies of GA
wsn) the inequality @k~(V,)@,,(l$) 2 0, V , E holds, and GB, respectively, then the maximum entropy of GA and
then nonconservation of entropy and ectropy in the isolated
large-scale dynamical system G implies Axiom i i ) . The GB individually is achieved at = e e and TB = %e,
above inequality postulates that the direction of energy flow respectively, while the maximum entro y of the composite
for any pair of energy similar subsystems is consistent; that system GC is achieved at Vzc = d4 * + qe~ ,. wiii)
, It follows
is, if for a given pair of connected subsystems at a given from Theorem 3.4 that for a stable equilibrium state V, =
energy level the energy flows in a certain direction, then e , where p 2 0 is the total energy of the system G and
for any other pair of connected subsystems with the same is the number of subsystems of 8, the entro y is totally
energy level, the energy flow direction is consistent with
the original pair of subsystems. Note that this assumption
B-
defined by p and p; that is, S(V,)= q log,(c+ ) plog, c.
does not specify the duection of energy flow between Dual criteria to the eight criteria outlined above can also be
subsystems. To see that S(V,(t))2 0, t 2 t o ,along with established for an analytic expression representing system
Axiom i i i ) implies Axiom i i ) note that since (57) holds for ectropy.
all t 2 to and V,(t,) E is arbitrary, (57) implies
Iv. TEMPERATURE
EQUIPARTITION
AND BOLTZMANN'S
KINETICTHEORYOF GASES
The thermodynamic axioms introduced in Section 111pos-
Now, it follows from (58) that for any fixed system energy tulate that subsystem energies are synonymous to subsystem
level V. E there exists at least one pair of connected temperatures. In this section, we generalize the results of
Section I11 to the case where the subsystem energies are
subsystems G k and GI, k # 1, such,that @ k ~ ( K ) ( w ~ i - u ~ k2)
proportional to the subsystem temperatures with the propor-
0. Thus, if wsk 2 v.1 (respectively, w,k 5 wsl), then tionality constants representing the subsystem specific heats.
&(V,) 5 0 (respectively, @ k l ( V , ) 2 0). Furthermore, In the case where the specific heats of all the subsystems
it follows from Axiom iii) that for any other pair of are equal the results of this section specialize to those of
connected subsystems 8, and &, m # n, with w,, 2 w,, Section 111. To include temperature notions in our large-
(respectively, ugm 5 W~,,) the inequality @mn(Vs) 5 0 scale dynamical system model we replace Axioms i) and
(respectively, &,,(V,) 2 0) holds which implies that i i ) of Section 111 by the following axioms. Let pi > 0,
@ m n ( V , ) ( ~ s n- wsm) 2 0, m # n. (59) i = 1,. . . , q, denote the reciprocal of the specific heat of
the ith subsystem G; so that the (empirical) temperature
Thus, it follows from (59) that energy (heat) flows naturally in ith subsystem is given by 7;' = p;v,i. Axiom i): For
from more energetic subsystems (hot objects) to less ener- the connectivity matrix C E WX'Jassociated with the
getic subsystems (cooler objects). Of course, since in the large-scale dynamical system G defined by (12) and (13),
isolated large-scale dynamical system G ectropy decreases if rank C = q - 1 and for = 1, i # j , Qij(V,) = 0
and only if entropy increases, the same result can be arrived if and only if fliw.; = pjwSj. Axiom ii): For i , j =
at by considering the ectropy of G. Since Axiom ii) holds, it
follows from the conservation of energy and the fact that the 1,..., q, (pjwsi - ppsj)@ij(V,), 5 0, V, E Axiom i )
large-scale dynamical system G is strongly connected that implies that if the temperatures in the connecteJsubsystems
nonconservation of entropy and ectropy necessarily implies Gi and Gj are equal, then heat exchange between these
energy equipartition. subsystems is not possible. This is a statement of the zeroth
law of thermodynamics which postulates that temperature
Finally, we close this section by showing that our def- equality is a necessary and sufficient condition for thermal
inition of entropy given by (40) satisfies the eight crite- equilibrium. Axiom i i ) implies that heat (energy) must flow
ria established in [45] for the acceptance of an analytic in the direction of lower temperatures. This is a statement
expression for representing a system entropy function. In of the second law of thermodynamics which states that a
particular, note that for a dynamical system G: i) S(V,) transformation whose only final result is to transfer heat
is well defined for every state V. E $+ as long as c > from a body at a given temperature to a body at a higher
0. i i ) If G is isolated, then S(V,(t))is a nondecreasing temperature is impossible. Next, in light of our modified
function of time. i i i ) If S,(u.,) = log,(c +
wsj) - log, c axioms we present a generalized definition for the entropy
is the entrop of the ith subsystem of the system G, then and ectropy of 8 . The following proposition is needed for
xi=,
S(V,) = B Si(vSi)= eTlog,(ce + V.)- qlog,c and the statement of the main results of this section.
408
 Pmposition 4.1: Consider the large-scale dynamical sys-
tem 8 with the power balance equation (6,and assume that
Axiom ii) holds. Then for all V,, E , :W tr 2 t o , and
S(t),t E [to,tf], such that K(tr) = KO,

and
.,. 0

where V.(-T) = 0 and &(to) = KOE E:.

where V.(t), t 2 to, is the solution to (6) with initial Theorem 4.1: Consider the large-scale dynamical system
~. . = KO.
condition K(to) 8 with the power balance equation (6) and assume that
Proof. The proof is identical to the proof of Propositions Axiom ii) holds. Then there exists an entropy and an
3.2 and 3.5. U ectropy function for 8. Moreover, S,(V,), V, E Et, and
S,(V,), V, E E !, are possible entropy functions for 8
Note that with the modified Axiom i) the isolated large- with S,(O) = S,(O)' = 0, and &,(V,), V , E and El,
scale dynamical system 8 has equilibrium energy states &,(V,), V, E g+,are passible ectropy functions for 8
given by V,, = ap, for (Y 2 0, wherep A [1/81,...,l/&IT. with €,(a) = &,(a) = 0. Finally, all entropy functions
As in Section 111, we define a reversible process as a process s(v~), V. E E:, for 8 satisfy
where the trajectory of the system 8 moves along the set
of equilibria for the isolated system 8 and an irreversible
process as a process that is not reversible. Thus, it follows &(K) 5 S(K) - S(0) i Se(V.), V , E E:, (68)
from Axioms i) and ii) that inequalities (60) and (61) are
satisfied as equalities for a reversible process and as strict and all ectropy functions E ( % ) , V. E E:, for 8 satisfy
inequalities for for an irreversible process.
Definition 4.1: For the large-scale dynamical system 8
Ea(%) 5 E ( & ) - E ( 0 ) 5 K E E:. (69)
Proof. The proof is identical to the proof of Theorems
with the power balance equation (6), a function S : Et i
W satisfying 3.1 and 3.3. 0

For the statement of the next result define p %

[l/Pl, ", l/p,]T and P 4 diag[@~,. . ,&].
Proposition 4.2: Consider the large-scale dynamical sys-
tem 8 with the power balance equation (6) and assume that
Axiom ii) holds. Then the function S : E t i R given by
for any t z 2 tl Z to and S(t), t E [tl,tz], is called the
entropy of 8.
S(K) = pTlog,(ce + PV.) - eTplog, c, K E E:, (70)
Definition 4.2: For the large-scale dynamical system 8
with the power balance equation (6), a function E : E t .+ R +
where log,(ce PV,) denotes the vector natural logarithm
satisfying given by [log,(c+Plv,~),..., I o g , ( ~ + P , u , , ) ] ~is,an entropy
function of 5.Furthennore, the function E :E: --t R given
&(V,(tz)) 5 &(V.(h)) by
+ 1: k P i u s i ( t ) [ s i ( t ) - uii(V,(t))]dt, (63)
>=1
E ( V , ) = $KTPK,
is an ectropy function of 8.
v. E E:, (71)
~~

for any t z 2 ti 2 to and S ( t ) ,t E [tl,tz],is called the Proof. The proof is identical to the proof of Propositions
ectmpy of 8. 3.4 and 3.6. 0
For the next result define
Remark 4.1: As in Section 111, it can be shown that
&(KO) a the entropy and ectropy functions for 8 defined by (70)
and (71) satisfy, respectively, (62) and (63) as equalities
for a reversible process and as strict inequalities for an
irreversible process.
409
 Once again, inequality (62) is Clausius' inequality for and be given by (71) and (70), respectively, and define
reversible and irreversible thermodynamics; while inequal- D, {Vs E : eTV, = p } , where 0 2 0 . Then,
ity (63) is an anti Clausius inequality. Moreover, for the
ectropy function given by (71) inequality (63) shows that a ar min(E(V,)) = ar$max(S(V,)) =
P
= -p. (73)
thermodynamically consistent large-scale dynamical system $El% .ED, eTP
model is dissipative with respect to the supply rate x T P S
and with storage function corresponding to the system Furthermore, Emin4 E(V,') = $&and&,, 6 S(V;) =
ectropy &(If,). In addition, if we let dQi(t), [ s ; ( t ) - e*pIog,(c+ &)- eTpIog,c.
uii(V.(t))]dt, i = 1,. . . , q , denote the infinitesunal amount
of heat received or dissipated by the ith subsystem of G Proof. The proof is identical to the proof of Proposition
over the infinitesimal time interval dt at the absolute ith 3.7 and hence is omined. U
subsystem temperature T; c + i3;u,;, then it follows from
(62) that the system entropy varies by an amount Proposition 4.3 shows that when all the energy of a
large-scale dynamical system is transformed into heat at
a uniform temperature, entropy is a maximum and ectropy
is a minimum.
Next, we provide a kinetic theory interpretation of the
Finally, note that the nonconservation of entropy and ec- (steady-state) expressions for entropy and ectropy presented
tropy equations (50) and (51), respectively, for isolated in this section. Specifically, we assume that each subsystem
large-scale dynamical systems also hold for the more gen- pi of the large-scale dynamical system B is a simple system
eral definitions of entropy and ectropy given in Definitions consisting of an ideal gas with rigid walls. Furthermore, we
4.1 and 4.2. The following theorem is a generalization of assume that all subsystems G; are divided by diathermal
Theorem 3.4. walls (i.e., walls that permit energy flow) and the overall
Theorem 4.2: Consider the large-scale dynamical system dynamical system is a closed system; that is, the system is
B with power balance equation (6) with S t ) 0 and separated from the environment by a rigid adiabatic wall. In
this case, 0;= k i n ; , i = 1 , . . . ,q, where n;, i = 1,.. , ,q,
d ( V , ) 0 and assume that Axioms i) and ih hzd. Then is the number of molecules in the ith subsystem and k > 0
for every 01 2 0, a p is a semistable equilibrium state of
(6). Furthermore, V,(t)i&peTV,(to) as t iM and is the Boltzmunn constant (i.e., gas constant per molecule).
Without loss of generality and for simplicity of exposition
&peTVs(to) is a semistable equilibrium state. Finally, if let k = 1. In addition, we assume that the molecules in
for some k E 11,. . . , q } , u k k ( K ) 2 0 and ukk(Vs) = 0 if the ideal gas are hard elastic spheres; that is, there are
and only if v.k = 0, then the zero solution Vs(t)= 0 to (6) no forces between the molecules except during collisions
is a globally asymptotically stable equilibrium state of (6). and the molecules are not deformed by collisions. Thus,
there is no internal potential energy and the system intemal
Proof. It follows from Axiom i) that a p E E :
, 01 2 0, energy of the ideal gas is entirely kinetic. Hence, in this
is an equilibrium state for (6). To show Lyapunov stability case, the temperature of each subsystem 0, is the average
of the equilibrium state a p consider the system shifted
ectropy &.(v,) = a(& -~ P ) ~ P- ( Vap) , as a Lyapunov
function candidate. Now, the proof follows as in the proof
translational kinetic energy per molecule which is consistent
with the kinetic theory of ideal gases.
of Theorem 3.4 by invoking-Axiom ii) and noting that Definition 4.3: For a given isolated large-scale dynami-
cal system G in thermal equilibrium define the equilibrium
h j ( V 8 ) = -@ji(Vd, v, E R .: i # j,i , j = 1,...I 4.
entropy of G by Se = nlog,(c + %) - nlog, c and
P p = e, and eTw(V,) = 0, V , E . :
E Alternatively, in
the case where for some k E {I, . . . , q } , u k k ( V , ) 2 0 and the equilibrium ectropy of 0 by E, = l(eTv.,)Z , where
akk(V.) = 0 if and only if v,k = 0, global asymptotic e T K , denotes the total steady-state energy of the large-
stability of the zero solution V,(t)= 0 to (6) follows from scale dynamical system 0 and n denotes the number of
standard Lya unov arguments using the system ectropy molecules in 8 .
E(Vs)= i V ,;PPV. as a candidate Lyapunov function. 0 Note that the equilibrium entropy and ectropy in Defi-
nition 4.3 is entirely consistent with the equilibrium (max-
It follows from Theorem 4.2 that the steady-state value imum) entropy and equilibrium (minimum) ectropy given
of the energy in each subsystem B; of the isolated large- by Proposition 4.3. Next, assume that each subsystem B; is
scale dynamical system G is given by V., = &peTV,(to) initially in thermal equilibrium. Furthermore, for each sub-
P system, let us; and n;, i = 1,. . . , q , denote the total internal
which implies that u s ; , = -pie
& - Pe T x ( t o ) or, equivalently,
energy and the number of molecules, respectively, in the
f';, = 0 ; ~= ~&eTV,(to). ; ~ Hence, the steady state ith subsystem. Hence, the entropy and ectropy of the ith
temperature Of the isorated large-scale dynamical system subsystem are given by S;= n; log,(c+v.,/n;) -%log, c
G given by T, = &eTV,(to)e is uniformly distributed and E; = $%, respectively. Next, note that the entropy and
over all the subsystems of 4. This phenomenon is known the ectropy of the overall system (after reaching a thermal
as temperature equipartition in which all the system energy equilibrium) are given by Se = nlog,(c+*)-nlog, c
is eventually transformed into heat at a uniform temperature
and hence all natural processes (system motions) would and E. = 'q. Now, it follows from the convexity
cease. of - log,(.? and conservation, of energy that the entropy of
Proposition 4.3: Consider the large-scale dynamical sys-
G at thermal equilibrium is given by
tem D with power balance equation (6), let E : -+ E+
and S : Rg",iR denote the ectropy and entropy of D
410
 (77) that for each subsystem the power balance equation
satisfies
4
% ( t )+Wu,i(t) + U<j[Wdt)- w&)] = s;(t)(79)
j=l,j#l

for all t 2 0. Note that &(V.) h ~ ~ = l , j + u i j [ -

= Ca. (74)
vsjji, V, E Ha,, i = 1, ...,q, represents the energy flow
w8;

i=l
from the ith subsystem to all other subsystems and is
Furthermore, the ectropy of G at thermal equilibrium is given by the sum of the individual energy flows from the
given by ith subsystem to the jth subsystem. Furthermore, these
energy flows are proportional to the energy differences
of the subsystems; that is, wsi - wsj. Hence, (79) is a
power balance equation that govems the energy exchange
among coupled subsystems and is completely analogous to
the equations of thermal transfer with subsystem energies
playing the role of temperatures. Furthermore, note that
since uij 2 0, i # j , i, j = 1,...,q, energy flows from more
energetic subsystems to less energetic subsystems which is
consistent with the second law of thermodynamics which
requires that heat (energy) must flow in the direction of

= x&i.
i=l
(75)
lower temperatures.
The next proposition is needed for developing expres-
sions for steady-state energy distributions of the large-scale
dynamical system G with linear power balance equation
It follows from (74) (respectively, (75)) that the equilib-
rium entropy (respectively, ectropy) of the system (gas) (77).
0 is always greater (respectively, less) than the sum of Proposition 5.1 (1241): Consider the large-scale dy-
entropies (respectively, ectropies) of the individual subsys- namical system G with power balance equation given by
tems G;. Hence, the entropy (respectively, ectropy) of the (77). Suppose KO 22 0 and S ( t ) 22 0, t 2 0. Then the
gas increases (respectively, decreases) as a more evenly solution Vp(t), t 2 0, to (77) is nonnegative for all t 2 0 if
distributed (disordered) state is reached. Finally, note that and only if W IS essentially nonnegative.
it follows from (74) and (75) that Se = CB=,Si and Next, we develop expressions for the steady-state energy
E, = CY=IEi ifand only if 2 ?,= i # j , i , j = 1,...,q; distribution for the large-scale dynamical system L2 for
that is, the initial temperatures of a d subsystems are equal. the cases where supplied system power S ( t ) is a periodic
function with period T > 0; that is, S ( t + T ) = S(t),t 2 0,
and S ( t ) is constant; that is, S ( t ) = S . Define e ( t ) 2
V. THERMODYNAMIC +
SYSTEMS WITH LINEARENERGY V,(t) - V.(t T ) , t 2 0, and note that
EXCHANGE e(t) = W e @ ) , e(0) = V,(O) - V.(T), t 2 0. (80)
In this section we specialize the results of Section III Hence, since
to the case of linear energy exchange between subsystems;
that is, u i j ( K ) = u,jv,., uij 2 0 , i , j = 1,...,q. In this e(t) = ewi[K(o)- X ( T ) ~ ,t 2 0 , (81)
case, the vector form of the energy balance equation (2), and W is semistable, it follows from iw) of Lemma 2 of
with t o = 0, is given by [23] that
lim e ( t ) =
t-m
lim [V,(t)- V,(t
t-m
+ T)]

= (Iq - WW#)[X(O)- V,(7)1, (82)

which represents a constant offset to the steady-state error
or, in power balance form, energy distribution in the large-scale dynamical system G.
For the case where S ( t ) s S, T + M and hence the
K(t) = W W )+ S ( t ) , K(0) = KO, t 2 0 , (77) following result is immediate. This result first appeared in
~31.
where W E Rqxq is such that
Proposition 5.2: Consider the large-scale dynamical sys-
tem G with power balance equation given by (77). Suppose
that V,, 22 0 and S ( t ) S 22 0. Then V,, 4
limi-, K ( t ) exists if and only if S E R ( W ) .In this case,
Note that (78) implies W(;,?)= -ujj 5 0 , j ,= v,, = (rq- WW#)V.,- W # S
1,...,q, and hence W is a semistable compartmental matnx. (83)
If ui; > 0 , i = 1,..., q, then W is an asymptotically and K, 22 0. If, in addition, W is nonsingular, then V , ,
stable compartmental matrix. An important special case of exists for all S 2> 0 and is given by
(77) is the case where W is symmetric or, equivalently,
c..-- uji,a # j , a, j = 1,..., 9. In this case, it follows from v,, = -w-'s. (84)
411
 Proof. Note that it follows from Lagrange's formula that the point? b [zl, ..., z,lT E V and time instant t 2 t o , let
the solution V,(t),t 2 0, to (77) is given by 4 : V x R+ x R" iR" denote the system energy flow
within the continuum V ; that is, d(x,v(x,t ) ,Vw(z, t ) ) =
V ( t )= eWtVso + 1 t
ew(t-")S(s)ds, t 2 0. (85) [41(z,v(z,t),V v ( z , t ) ) ,...,4"(z,w(z,t),vw(z,t))lT,
where &(., ., .) denotes the energy flow through a unit
area per unit time in the x i direction for all i = 1,...,n
Now, the result is a direct consequence of Proposition 5.1
and iw), viz), wiii), and iz) of Lemma 2 of [23]. and Vw(z,t) [ D i ~ ( z , t..., ) , D , ~ ( z , t ) ] z, E V , t 2 t o ,
U denotes the gradient of w(.,dwith respect to the spatial
variable z, and let s : V x R+ + K, denote the energy
Next, we specialize the result of Proposition 5.2 to (heat) flow into a unit volume per u t time from sources
the case where there is no energy dissipation from each uniformly distributed over V . Hence, a power balance
subsystem B; of G; that is, uii = 0, i = 1,...,q. Note that in equation over a unit volume within the continuum V
this case eTW = 0 and hence rank W 5 q- 1. Furthermore, involving the rate of energy density change, the extemal
if S = 0 it follows from (77) that eTx(t) =, e T W K ( t ) = supplied power (heat flux), and the energy (heat) flow
0, t 2 0, and hence the total energy of the isolated large- within the continuum yields
scale dynamical system G is conserved.
Pmposition 5.3: Consider the large-scale dynamical sys- -aw(z,t)
= - V ' ~ ( z , v ( z , t ) , V v ( z , t )+)S ( Z , t ) ,
tem with power balance equation given by (77). Assume at
r a n k W = q - l , a i ; = O , i = l ,_..,q, and We = 0. If t 2 t o , (89)
V,, 22 0 and S = 0, then the steady-state energy distribu- tJ(z,to) = wo(z), 2 E V ,
tion V. of the isolated large-scale dynamical system G is $(z, v(x,t ) ,Vv(z, t ) ) ' h ( 2 ) 2 0, 2 f av,
t 2 to, (90)
given by
where V denotes the nabla operator, A(z) denotes the
outward normal vector to the boundary OV (at z) of the set
V , "." denotes the dot product in R", and wo(.) E X is a
given initial energy density distribution. The power balance
Proof. The proof is similar to the proof of Theorem 3.4 (conservation) equation (89) describes the time evolution
with .I(%) = WV,. U of the energy density w(z,t)over the region V while the
boundary condition in (90) involving the dot product implies
that the energy of the system G can either be stored or
Finally, we examine the steady-state energy distribution dissipated but not supplied through the boundary of V .
for the large-scale dynamical system G in case of strong Furthermore, we denote the energy density distribution over
coupling between subsystems; that is, u;j + 00, i # j . For the set V at time t 2 t o by ut E X so that for each
this analysis we assume that W given by (78) is symmetric; t 2 t o the set of mappings generated by wt(z) v(z, t )
that is, a,j = a.,, i # j , i , j = 1,...,q. and a,, > 0, i = for every z E V gives the flow of G. We assume that the
1, ...,q. Thus, -b is a nonsingular M-matrix for all values function $(., ., .) is continuously differentiable so that (89),
of a;,, a # j , = 1,...,q. Moreover, in this case it can (90) admits a unique solution w(z,t), 2 E V , t 2 to, and
be shown that if 2 -+ 1 as uij -+ 00, i # j , and ukl -+ ~ ( , , tE) X , t 2 to, is continuously dependent on initial
M, k # 1, then energy density distribution vo(z), z E V . It is well known
that if (89) is strictly parabolic, and WO(.) is a C2 function
with com act support and its derivative is sufficiently small
on [to,mp,then the classical solution to (89), (90) breaks
down at a finite tune. As a consequence of this, one may
Hence, in the limit of strong coupling the steady-state only hope to find generalized (or weak) solutions to (89),
energy distribution V, , given by (84) becomes (90) Over the semi-infinite interval [to,ca),that is, L,
functions U ( . , .) that satisfy (89) in the sense of distributions.
As in Section 111, to ensure a thermodynamically con-
sistent energy flow infinite dimensional model we require
which implies energy equipartition. This result first ap- the following axioms analogous to Axioms i) and ii).
peared in [23]. Axiom i)': For every 1: E V and unit vector U E R",
4(.,wt(z),Vvt(z)) U = 0 if and only if Vv,(z)u = 0.
Axiom ii)': For every z E V and unit vector U E R",
VI. CONTINUUM THERMODYNAMICS 4(z, wt(z),Vwt(z)) U > 0 if and only if Vvt(z)u < 0,
and ~(z,wt(z),Vwt(z))~u < 0 i f a n d o n l y i f V v t ( z ) u > 0.
In this section we extend the results of Section 111 to the Note that Axiom i)' implies that 4 i ( z , t ~ ~ ( z ) , V v ~ (=z 0) )
case of continuous thermodynamic systems wherein the if and only if Divt(z) = 0, z E V , i = 1,.._,n, while
subsystems are uniformly distributed over an n dimensional Axiom ii)' implies that Qi(z,wt(z), Vvt(z))D;v!(z) I
space. Since these systems involve distributed subsystems 0, z E V , i = 1,...,n, which further implies that
they are described by partial differential equations and Vvt(z)d(z, v t ( z ) ,Os(.)) _< 0, z E V ; that is, energy
hence are infinite dimensional systems. Specifically, we (heat) flows from regions of higher to lower energy densi-
consider continuous dynamical systems B defined over a ties. If s(z,t ) = 0, then Axioms i)' and ii)' along with the
compact connected set V c W" with a smooth (at least C') fact that $(z, w(z,t), Vw(z,t ) ) .A(!) 2 0, z E a V , t ? to,
boundary 8V and volume VL,). Furthermore, let X denote imply that at a given instant of tune the energy of the
a space of two-times continuously differentiable scalar dynamical system B can only be transported, stored, or
functionsdefinedonV,letv(z,t),wherev:V x R + i R + , dissipated but not created. Next, we establish the classical
denote the energy density of the dynamical system D at Clausius inequality for our thermodynamically consistent
412
infinite dimensional energy flow model given by (89), and c > 0, is called the entropy of G.
(90). For the remainder of this section dV represents an Theorem 6.1: Consider the dynamical system 8 with the
infinitesimal volume element of V , SV denotes the surface power balance equation (89), (90) and assume that Axiom
enclosing V and dSv denotes an infinitesimal boundary ii)' holds. Then the function S : X + W given by
element.
Pmposition 6.1: Consider the dynamical system 8 with
the power balance equation (89), (90) and assume that
S(4= J, k ( c + 9tb))dV - Vvollog, c, (95)
Axiom ii)' holds. Then, for every initial energy density
distribution VO(.) E X,tf 2 to, and s ( t ) , t E [to,tf],such is an entropy function for 8.
that ut,(.) = vo(z), Proof. It follows from the Green-Gauss theorem, Axiom
ii)', and (95) that

2 dtl. (96)
Now, integrating (96) over [ t l ,121 yields (93). 0

Next, we establish a dual inequality to inequality (91)

that is satisfied for our thermodynamically consistent energy
flow model.
Proposition 6.2: Consider the dynamical system B with
the power balance equation (89), (90) and assume that
Axiom ii)' holds. Then, for every initial energy density
distribution 2r0(.) E X,tf 2 to, and s ( t ) , t E [to,tf],such
that vt,(z) vo(z),

2 0, (97)
where v(z, t ) , 2: E V ,t 2 t o , is the solution to (89), (90).
5 0, (92) Proof. It follows from the Green-Gauss theorem and
which proves the result. 0 Axiom ii)' that

Next, we give the entropy definition for continuous

dynamical systems.
Definition 6.1: For the dynamical system 8 with the
power balance equation (89), (90), the function S : X + W
satisfying

S(%) 2 S(%) + 1; dt)dt, (93)

for all s ( t ) , t 2 t o , and t2 2 tl 2 to,where

413
 that infinitesimal increment in the entropy of 0 over the
infinitesimal time interval dt satisfies

-lo'lv Vu(z, t)$(z,U ( + , t ) ,Vw(z, t))dVdt

- 1;Lv 4 2 , t)4(z, ~ ( zt )
, ,W z , t ) ) fi(+)dSvdt

=- 1:J, Vv(z,t)4(z,v(z,t),Vv(z,t))dVdt
where the shifted energy density c+vt(z) plays the role of
(absolute) temperature at the spatial coordinate z and time
t. For an isolated dynamical system 0; that is, s(z,t ) E 0
2 0, (98) and $ ~ ( z , v ( z , t ) , V v ( z ,.A(z)
t ) ) z 0, z E ab', (93) and
which proves the result. 0 (99) yield the fundamental inequalities
S(vt,) 2 S(ut,), t z 2 tl, (104)
Definition 6.2: For the dynamical system 0 with the
power balance equation (89). (90), the function & : X --t R and
satisfying &(ut,) I &(ut,), t z 2 t l . (105)

&(ut,) I &(vt,)+ Vvo1 1; Wdt, (99)

Hence, for an isolated infinite dimensional system 4 the
entropy increases if and only if the ectropy decreases. It is
important to note that (105) also holds in the case where
for all s ( t ) , t 2 to, and t 2 2 t i 2 t o , where $(z, v(z, t ) ,Vw(z, t ) ) A(%) $ 0, z E OV, whereas (104)

@(t) 2 /V
~ ( zt ),s ( z ,t)dV
does not necessarily hold in that case.
The next theorem shows that the infinite dimensional
thermodynamic energy flow model has convergent flows
to Lyapunov stable uniform equilibrium energy density
- J,, 4 z , t ) 4 ( z d z 3 t ) ,V v ( z , t ) ) . fi(+)dSvXIOO) distributions determined by the system initial energy den-
sity distribution. However, since our continuous dynamical
is called the ectmpy of G.
system G is defined on the infinite dimensional space X,
Theorem 6.2: Consider the dynamical system G with the bounded orbits of G may not lie in a compact subset of
power balance equation (89), (90) and assume that Axiom X which is crucial to being able to invoke the invariance
ii)' holds. Then the function & : X 4R given by principle for infinite dimensional dynamical systems [46].
This is in contrast to the dynamical system 0 considered in
&(v,) = -vvol uz(z)dV, (101) the previous sections arising from a power balance (ordinary
2 -
differential) equation defined on a finite dimensional space
is an ectropy function for 4. wherein local boundedness of an orbit of P ensures
Proof. It follows from the Green-Gauss theorem, Axiom that the orbit belongs to a compact subset of z+.
Hence,
to ensure that bounded orbits of G lie in compact sets we
ii)', (90), and (101) that
construct a larger space H 3 X as a Sobolev space so
that by the Sobolev embedding theorem [47], [48] there
exists a Banach space B 3 H such that the unit ball in
H belongs to a compact set in B; that is, H is compactly
embedded in B. In this case, it follows from Lemma 3
of 1461 that a bounded orbit of the dynamical system 0
defined on H has a nonempty compact, connected invariant
omega limit set in 8.For the next result, the L2 operator
norm I( 1 1 on ~ X~ is used for the definitions of Lyapunov,
semi, and asymptotic stability. Furthermore, we introduce
the Soholev spaces W,'(V) {at : V -+ W : ut E
C'(V) n Lz(V), (Vwt)= E Lz(V)},, and W,(V) 4 {ut :
V -+ R : ut E C o ( V )n L2(V)},, = Lz(V), where
denotes completion of {.} in Lz in the sense of [48], with
norms

Inequality (93) is precisely Clausius' inequality for re-

versible and irreversible thennodvnamics as aoolied to
infinite dimensional systems; whife inequality (45) is an
ilutllW: ' l I V t l l L ~= [ / v:(z)dV11
LJV
' > (107)
anti Clausius inequality that shows that a thermodynami- defined on W,'(V) and Wg(V), respectively, where the
cally consistent infinite dimensional dynamical system is gradient Vvt(z) in (106) is interpreted in the sense of
dissipative with storage function corresponding to the sys- a generalized gradient [48]. Note that since the solutions
tem ectropy. In addition, note that it follows from (93) to (89), (90) arc assumed to be two-times continuously
414
differentiable functions on a com act set V, it follows that a E aV, that
P
wt(z),t 2 to,belongs to both W , ( V ) and W:(V).
Theorem 6.3: Consider the dynamical system 0 with
power balance equation (89), (90) with s ( z , t ) 3 0 and
@(z,w(z,t),Vu(z,t)) . A(z) = 0. z E aV. Assume that
Axioms i) , ii)’ hold and
a
V2vt(z)V. 4(.,%(.), --.t(.)) 5 0,
ax
z EV, wt E W,(V), (108)

where V2 4? V.V denotes the Laplacian operator. Then for

every a 2 0, u(x,t ) a is a semistable equilibrium state
of (89), (90). Furthermore, u(z,t) + &Jvwo(s)dV as
t -+ co for every initial energy density distribution WO(.) E
W$(V) and every I E V ; moreover, &
J, vo(a)dV is
a semistable equilibrium distribution state of (89), (90).
Finally, if s(z,t)= 0 and there exists at least one point
zp E aV such that 4(zprut(z),Vwt(z,)). jl(zp) > 0
and ~(zp,wt(zp),Vwt(zp)). fi[zp) = 0 if and only if
wt(zp)= 0, then the zero solution w(z,t)= 0 to (89),
(90) is a globally asymptotically stable equilibrium state of + V2+, t ) V .4 b , ~ ( 2$1,
, V+, t))dV,
(891, (90).
Proof. It follows from Axiom i)’ that w(z,t) 2 a,(I 2 0, (112)
is an equilibrium state for (89), (90). To show Lyapunov
stability of the equilibrium state w(z,t) = cy consider the where D6(z)w(z, e
t ) Vu(z, t)A(z) denotes the directional
system shifted ectropy &,(wt) = f Jv(ut(z)- a)*dV = derivative of w(z,t) along A(z) at z E aV. Next, note
qllvt - (111;~
1
as a Lyapunov function candidate. Now, it that for the isolated dynamical system 0 with the boundary
follows from the Green-Gauss theorem and Axiom ii)’ that condition 4(z, w(z,t ) ,Vw(z, t ) ) . fi(z) = 0, z E aV, it
follows from Axiom i)’, with U = A(z), that D,(,,w(z, t ) =
0, z E dV. Hence, it follows from Axiom ii)’, (108),
and (112) that El(ut) 5 0, t 2 to, for any u0(.) E
W i ( V ) .Furthermore, since the functions &I ( w t ) and &0(wt)
are nonincreasing and bounded from below by zero, it
follows that &l(ut) and &0(wt) are bounded functions for
every WO(,) E Wi(V). This implies that the positive orbit
r+(wo) {U(., t ) : I E V >t E [to, co)} of G is bounded
in Wj(V) for all wo(.) E W i ( V ) .Hence, since Wi(V) is
compactly embedded in W i ( V ) ,it follows from Sobolev’s
embedding theorem [47 481 that y+(uo) is contained in
d,
a compact subset of W,(V). [ Next, define the sets D;, =
{ ~ Et W:(V) : E i ( ~ t <) q } and Vw; = { ~ E t W,O(V):
&(ut) < q } for some arbitrary q > 0. Note that TIw,
and Dw: are invariant sets with respect to the dynamical
system G . Moreover, it follows from the definition of El ( u t )
Vu(z, t)4(z,V(z3t ) *VU(z,t))dV and &(ut) that Dw; and D,; are bounded sets in Wi(V)
and W j ( V ) ,respectively. Next, let R {ut E D,o :
5 0, (109)
io(%) = 0) = {ut E D,; : Vwt(z)$(z, ut(.), vwt(z);=
which establishes Lyapunov stability of the equilibrium 0, 2 E V } . Now, follows from Axioms i)’ and ii)’
state w(z,t ) 2 (I. that R = {ut E Dwp : Vwt(z) = 0, z E V } or
Next, to show semistability of this equilibrium state,
consider the following (scaled) ectropy and ectropy-like
Lyapunov functions
which are the equilibrium states of (89), (90). Since the set
&(Ut) = lIWtll2W,o> ut E W,o(V), (110) R consists of only the equilibrium states of (89), (90), it
follows that the largest invariant set M contained in R is
El(%) = lvtl12W;, ut E W : ( V ) . (111) given by M = ‘R. Hence, noting that M belongs to the set
of generallzed (or weak) solutions to (89), (90) defined on
It follows from (99) with s ( z , t ) 3 0 that &o(wt) is a R,it follows from Theorem 3 of [46] that for any initial
nonincreasing function of time for all vu(.) E W,O(V). energy density distribution uo(.) E D,;, u ( z ,t ) -+ M
Furthermore, it follows from the Green-Gauss theorem and as t + co with respect to the n o m 11 . Ilw; and hence
the boundary condition #(z, v(z,t ) ,Vu(., t ) ) fib) = 0, w(z,t ) G cy is a semistable equilibrium state of (89), (90).
415
Moreover, since 4 > 0 can be arbitrarily large but finite Remark 6.1; Condition (108) physically implies that for
and El(vt) is radially unbounded, the previous statement an energy density distribution vt(z),z E U , the energy
holds for all vo(.) E W:(V).Next, note that since, by the flow $ ( I ,wt(z),Vvt(z)) at z E V is proportional to the
energy density at this point. Note that for a linear ener
divergence theorem, Y
flow model; that is, $ ( z , v t ( ~ ) , V v t ( z=) )-k(Vvt(z)l ,
where k > 0 is a conductivity constant, condition (108) is
automatically satisfied with - k [ V 2 v t ( z ) ] 5
z 0, I E V .
Finally, we give an analogous proposition to Proposition
4 ( ~ , 4 1t ), ,V v ( z , t ) )f i ( ~ dSv
) 3.7 for infinite dimensional systems.
= 0, (113) Proposition 6.3: Consider the dynamical system P with
power balance equation (89), (go), let E : X + R and S :
it follows that J, v(z, t)dV = s,
uo(z)dV, t 2 to, which X iR denote the ectropy and entropy of P and be given
implies that v(z,t ) -, & s, vo(z)dV as t + CO. b (95) and (101), respectively, and define VcA {vt E X :
Finally, we show that if s(z, t ) = 0 and there exists at vt(z)dV = P } , where P 2 0. Then,
least one point z PE aV such that 4(z vt z p ) , V u t ( z p ) ) .
fi(zp)> 0 and 4(zp,vt(zp),Vvt(zp)~'. li[zp)= 0 if and argmin(&(vt))= argmax(S(vt)) = vt = -.
P (115)
only if vt(zp) = 0, then the zero solution v ( z , t ) E 0 to WED, "tEW. V"O1
(89), (90) is a globally asymptotically stable equilibrium
state. Note that it follows from the above analysis with Funhermore, &,in &(U;) = $
and S,,, S(vt) =
a = 0 that the zero solution v(z, t ) 0 is semistable 1.'"oIIlog,(c+ - log,cl.
and hence a Lyapunov stable equilibrium state of (89),
(90). Furthermore, it follows from Axiom ii)' that if Proof. The proof is similar to the proof of Proposition
D,(,p)v(zp,t) > 0 for zp E aV and some t 2 t o , 3.7 and hence is omitted.
then the energy density decreases at this point; that is, We close this section by noting that the results of this
8.J(= J ) section can be easily generalized to the case where the
-;if < 0 and D , q z p ~ v ( z p , < t ) 0.
~ Altema- energy density at a p o i n t I E V is proportional to the
tively, if DiL(=) u ( z p , t )< 0, then > 0 and temperature; that is, T ( z ,t ) = P(z)v(s,t ) ,where T ( z ,t ) is
D n ( z , ) v ( z p , t ) W < 0. Thus, it follows from Axiom the (empirical) temperature distribution over the continuum
ii)', (108), and (1 12) that h ( v t )is a nonincreasing function and P(z) is the reciprocal of the specific heat at the spatial
of time for all vo(.) E W,l(V)and since E l ( v t ) is bounded coordinate 2 . In this case, analogous results to the results of
from below by zero, the positive orbit yf(vo) of G is Section IV can be easily derived for the infinite dimensional
bounded in W i ( V ) .Hence, since W,'(V) is compactly thermodynamic model. Finally, it is important to note
embedded in W i ( V )it follows from Sobolev's embedding that the results of this section apply to an arbitrary (not
theorem [47], 1481 that y+(vo) is contained in a compact necessarily Cartisian) n-dimensional space. In particular,
subset of W,"(U).Next, consider the (scaled) ectropy Lya- we could consider a coordinate transformation y = Y ( z ) ,
Y ( 0 )= 0 and Y : U * W n is a diffeomorphism in
punov function &(ut) and note that the Lyapunov derivative where
is given by the neighborhood of the origin, so that y is defined on the
image of V c R" under Y . In this case however, the nabla
and gradient operators need to be redefined appropriately.

VII. CONCLUSION
In this paper we have attempted to outline a general sys-
tem theory framework for thermodynamics. The proposed
macroscopic mathematical model is based on a nonlinear
(finite and infinite dimensional) compartmental dynamical
system model that is characterized by energy conservation
laws capturing the exchange of energy between coupled
-
i0. (114) macroscopic subsystems. Specifically, using a large-scale
systems perspective, we developed some of the fundamental
Fulthermore, let R S? {ut E Dw; : &0(vt)= 0) = [ut E properties of irreversible thermodynamic systems involving
-
DW; : Vvt(z)4(z,vt(z), VV,(S)) E 0, I E V } n {vt E conservation of energy, nonconservation of entropy and
t ( ~ ) )= 0 , I E 8V). NOW, ectropy, and energy equipartition. This model is formulated
Vw; : 4 ( ~ , ~ ( z ) , V z i .A(z)
since Axioms i)' and ei)' hold, R = {ut E Dw; : and in the language of dynamical systems and control theory
it is argued that it offers conceptual advantages for
vvt(2) = 0 , I E V } n {vt E Dw: : = describing general thermodynamic systems.
0 for some zp E a V } = {0} and the largest invariant set The underlying intension of this paper has been to present
M contained in R is given by M = {O}. Hence, it follows one of the most useful and general physical branch of
from Theorem 3 of [46] that for any initial energy density science in the language of dynamical systems theory. The
distribution vo(.) E Dw;,v(z,t)+ M = {O} as t -t 00 laws of thermodynamics reign supreme among the laws of
with respect to the norm /I llwp which, since 1) > 0 is Nature and it is hoped that this paper will help to stimulate
arbitrary and &I (ut) is radially unbounded, proves global increased interaction between physicists and dynamical sys-
asymptotic stability of the zero equilibrium state of (89), tems and control theorists. Besides the fact that irreversible
(90). 0 thermodynamics plays a critical role in the understanding of
416
ou expanding universe, i t forms the underpinning of sevcral [21] M. Pawn, “Stochastic control and nonequilibrium thermodynamical
fundamental life science and engineering disciplines includ- sys1ems;’Appl. Molh. Oplim., vol. 19, pp. 187-202, 1989.
ing biological, physiological, and phsrmocological systems, [22] J. Brunet, “Information lheary and thermodynamics,” Cyhemefico,
chemical reaction systems. queuing systems. ecoloyicsl vol. 32, pp. 45-78, 1989.
I231 D. S . Bemstein and D. C. Hyland, “Compartmental modeling and
systems, demographic systems, telecommuniiatidn systems, second-moment analysis of stale space sysl“:’ S I A M J Moirk
transportation systems, network systems, and power systems Anal. Appl., vol. 14, pp. 880-901, 1993.
to cite but a few examples. [24] W.M. Haddad, V. Chellaboina, and E. August, “Stability and dissipa-
Finally, future urork will involve rystem-theoretic formu- tivity theory for nonnegative dynamical systems: A thermodynamic
lations of microscopic theorics of irreversible thcrmody- framework for biological and physiological systems; in Pmc. IEEE
Conf Dec. Conrr, Orlando, FL, 2001. pp. 442458.
namics and noncquilibrium StAtisticaI mechsnics and sta-
[25] R. W. Bracken and 1. C. Willems, “Stochastic c ~ n t r oand
l the second
tistical quantum mechanics. The newly devsloped notion law of thermodynamics,” in Pmc. IEEE CmJ Dec. Conrr, San
o f ectropy proposed in this paper involving an analytical Diego, CA, 1978, pp. 1007-1011.
description of an objective property of matter cnn potentially [26] D. S . Uemstein and S. P. Bhat, “Energy equipanition and the
ofler a conceptual advantage o w ]he beveral suhjectivr: emergence of damping in lossless systems,” in Pme JEEE Con/
quantum expressions for entropy proposed in the literature Dec. Conrr, Las -$egas, W ,2002, pp. 2913-2918.
(e.g., Darbczy entropy, Hanky entropy, Rcnyi cntropy, von [27] I. C. Willems, ”Dissipative dynamical systems part I: General the-
Neumsnn entropy, intinitc-norm entropy) involving a nie3- ory,” Arch. Rolionol Mech. Anal., vol. 45. pp. 321-351, 1972.
sure of information. An even more imponant benefit of the [28] A. G w e n , G. Keller, and G. Wamecke, Eds., Enlmpy. Princeton,
dynamical system represenlation of thermodynamics is the NI: Princeton University Press, 2003.
[29] R. H. Lyon, Slr?lislicnl Energv Analysis o/ Dynamical Sy~lemr:
potential of developing a unified classtc~lor quantum themy Theow ond Applications. Cambridge, MA: MIT Press, 1975.
which encompasses both niechnnics and thermodynaniics [30] P, W. Smith, ”Statistical models of coupled dynamical systems
without thc nced for ,tatistical (subjective or informa~ional) and the, transition from weak to strong coupling:’ Joumd of the
prohabilttiss Aeouslicol Sociely of America, vol. 65, pp. 695-698, 1979.
1311 J. Woodhouse, “An approach to the theoretical background of tati is-
tical energy analysis applied l o ~fructuralvibration,” Jounrol of Ihe
REFERENCES Acousricol Sociely o/America, vol. 69, pp. 1695-1709, 1981.
[32] A. J. Keane and W. G. Wce, “Statistical energy analysis ofstrongly
[I] R. G. Sachs, The Physics of Time Reversol. Chicago, IL: University coupled systems,” J Sound Yibr, vol. 117, pp. 363-386, 1987.
of Chicago Press, 1987. [33] R. S. Langley, “A general derivation of the statistical energy analysis
[2] H. D. Zeh, The Physical Basis o/lhe Direclion of Time. New York, equations for coupled dynamic systems:’ J. Sound Yibc, vol. 135,
S~rinw-Verlae.1989. pp. 499-508. 1989.
[3] M. C.-Maikey, I%’s Armw: The Origins of Thermodynamic Be- [34] A. Carcatem, “An entropy formulation for lhe analysis of cncrgy
havior. New York, N F Springer-Verlag. 1992. flow between mechanical resonators:’ Mechonicol Sysrems and Sig-
141 E. F. Oben, Concepu of Themrodynomics. New York, N E McGraw- no1 Pmcessing. vol. 16, pp. 905-920, 2002.
Hill, 1960. (351 Y. Kishimato and D. S . Bemstein, “Thermodynamic modeling of
[5] L. Tisza. in A Crilicol Review of Thennodynamics, E. B. Stuart, interconnected systems I: Conservative coupling,”J Somd Ehz, vol.
B. Gal-Or, and A. J. Brainard, Eds. Baltimore: Mono, 1970. 182, pp. 23-58, 1995.
[6] D. S. L. Cardwell, Fmm Wall lo Clausius: The Rise of Thermody- [36] -, “Thermodynamic modeling of interconnected syslems I 1 Dis-
nomic8 in the E& lndusrriol Age. Ilhaca, N Y Cornell University sipative coupling.” J. Sound Yibz, vol. 182, pp. 59-76, 1995.
Press, 1971. [37] A. Berman and R. J. Plemmons, Nomegorive Molricece~ in the
[7] S . G. Brush, The Kied o/ Molion We Col1 Heal; A H i m r y o/ lhe MorhemalicolSciencer. New York, Ny. Academic Presa, Inc., 1979.
Kinelic Theory in lhe Nineteen Cenlury. Amsterdam: North Holland, [38] D. S. Bemstein and S . P. Bhat, “Nonnegativity, reducibilty and
1976. semistability of mass action kinetics:’ in Pmc. IEEE Conf Dec.
[81 C. Truesdell, The Tragicomical H i m ? of Thermodynvnomics 1822- Conlx, Phoenix, AZ, 1999, pp. 2206-221 1.
1854. New York, N Y Springer-Verlag. 1980. [39] . IMeixner, “On the foundation of thermodynamics of processes:’ in
.
[9] P. Coveney, The A m w of 7ime. New York, N F Ballantine Books, A Criricol Review of TZhermodynomics, E. U. Stuart, B. Gal-Or, and
1990. A. 1. Brsinard, Eds. Mono Book Corp., pp. 37-47, 1970.
[IO] E. P. Gyfiopoulos and G. P. Beretta, Thermodynomics: Foundorions [40] D. J. Hill and P. I. Moylan, “Dissipative dynamical systems: Basic
and Applications. New York, NY Macmillan, 1991 input-output and stale propenies:’ J Franklin Insilute, vol. 309, pp.
[II] M. Goldstein and 1. F. Goldstein, The Refrigwaror andihe Universe. 327-357, 1980.
Cambridge, MA: Harvard University Press, 1993. [41] H. K. Khalil, Nonlineor Systems. Upper Saddle River, NI: Prentice-
[I21 H. C. Vas Baeyer, Menvellk Demon: Why Warmth Disperses ond Hall, 1996.
Time Passer. New York, Ny: Random House, 1998. [42] R. K. Pearson and T. L. Johnran, “Energy equipanition and
1131 U. D. Coleman and W. Noll. “The thermodynamics of elastic Ructuation-dissipation theorems for damped flexible structures:’
matrrials wilh heat conduction and viscosity:’ Arch. Ration. Mech. Quuarl. Appl. Marh., vol. 45, pp. 223-238, 1987.
Anolysb, vol. 13, pp. 167-178, 1963. [43] S . R. Hall, D. G. MacMartin, and D. S . Uemstein, “Covariance
(141 1. MUer, “Die Kiltefunktion, eine universelle Funktion in der Ther- averaging in Ihe analysis of uncertain systems:’ in Pme. IEEE Conj
modynamik vismer wi(rme1eitender Fliissigkeiten,” Arch. Rarion. Dec. Conlr, Tucson, AZ, 1992, pp. 1842-1859.
Mech. Andyds. vol. 40, pp. 1-36, 1971. [44] I. W,Gibbs, The Scientr/re Popem o/ J Willard Gibbs, Vol. I
I151 C. Truesdell, Rorionol Thermodymmics. New York, NY: McGraw- Thermodynamics. London: Longmans. 1906.
Hill, 1969. [45] E. P. Gyflopoulos and E. Cubukcu, “Enbopy: Thermodynamic def-
[I61 B. D. Coleman, “The thermodynamics of materials with memory:’ inition and quantum expression:’ Phys. Rev. E. vol. 5 5 , no. 4, pp.
Amh. Ration. Mech. Analysis, vol. 17. pp. 1 4 6 , 1964. 3851-3858, 1997.
1171 M. Gunis, “On the thermodynamics of materials with memory,” [46] 1. K. Hale, “Dynamical systems and stability,” 3. Moth. Analysis ond
Arch. Rorion. Mech. Analysis. vol. 28. pp. 4r350. 1968. Appl., vol. 26, pp. 39-59, 1969.
[I81 W. A. Day, “Thermodynamics based on a work axi0m:’Arch. Rorion. [47] S. L. Sobolev, “Applications of functional analysis in mathematical
Mech. Anolysis, vol. 31, pp. 1-34. 1968. physics,” Tmnslarions of Molhemorical Monogrophs. Am. Marh. Soc.,
f191 -, % theory of thermodynamics for materials with memory; vol. 7, 1963.
Arch. Rolion. Mech. Andysis, vol. 34, pp. 8 6 9 6 , 1969. [48] L. R. Volevich and B. P. Pancyakh, “Certain spaces of generalized
[20] 1. M. Ziman, Models o/Disorder. Cambridge, England: Cambridge functions and embedding lheorems:’ Russion M O L . Survqvs, vol. 20,
Univ. Press, 1979. pp. 1-73, 1965.

417